[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
#
|
|
|
|
# Licensed to the Apache Software Foundation (ASF) under one or more
|
|
|
|
# contributor license agreements. See the NOTICE file distributed with
|
|
|
|
# this work for additional information regarding copyright ownership.
|
|
|
|
# The ASF licenses this file to You under the Apache License, Version 2.0
|
|
|
|
# (the "License"); you may not use this file except in compliance with
|
|
|
|
# the License. You may obtain a copy of the License at
|
|
|
|
#
|
|
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
#
|
|
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
# See the License for the specific language governing permissions and
|
|
|
|
# limitations under the License.
|
|
|
|
#
|
|
|
|
|
|
|
|
"""
|
|
|
|
Package for distributed linear algebra.
|
|
|
|
"""
|
|
|
|
|
|
|
|
import sys
|
|
|
|
|
|
|
|
from py4j.java_gateway import JavaObject
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
from pyspark import RDD, since
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
from pyspark.mllib.common import callMLlibFunc, JavaModelWrapper
|
2017-05-03 04:58:05 -04:00
|
|
|
from pyspark.mllib.linalg import _convert_to_vector, DenseMatrix, Matrix, QRDecomposition
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
from pyspark.mllib.stat import MultivariateStatisticalSummary
|
2019-07-09 17:39:21 -04:00
|
|
|
from pyspark.sql import DataFrame
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
from pyspark.storagelevel import StorageLevel
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
|
2017-05-03 04:58:05 -04:00
|
|
|
__all__ = ['BlockMatrix', 'CoordinateMatrix', 'DistributedMatrix', 'IndexedRow',
|
|
|
|
'IndexedRowMatrix', 'MatrixEntry', 'RowMatrix', 'SingularValueDecomposition']
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
|
|
|
|
class DistributedMatrix(object):
|
|
|
|
"""
|
|
|
|
Represents a distributively stored matrix backed by one or
|
|
|
|
more RDDs.
|
|
|
|
|
|
|
|
"""
|
|
|
|
def numRows(self):
|
|
|
|
"""Get or compute the number of rows."""
|
|
|
|
raise NotImplementedError
|
|
|
|
|
|
|
|
def numCols(self):
|
|
|
|
"""Get or compute the number of cols."""
|
|
|
|
raise NotImplementedError
|
|
|
|
|
|
|
|
|
|
|
|
class RowMatrix(DistributedMatrix):
|
|
|
|
"""
|
|
|
|
Represents a row-oriented distributed Matrix with no meaningful
|
|
|
|
row indices.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
|
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
rows : :py:class:`pyspark.RDD` or :py:class:`pyspark.sql.DataFrame`
|
|
|
|
An RDD or DataFrame of vectors. If a DataFrame is provided, it must have a single
|
|
|
|
vector typed column.
|
|
|
|
numRows : int, optional
|
|
|
|
Number of rows in the matrix. A non-positive
|
|
|
|
value means unknown, at which point the number
|
|
|
|
of rows will be determined by the number of
|
|
|
|
records in the `rows` RDD.
|
|
|
|
numCols : int, optional
|
|
|
|
Number of columns in the matrix. A non-positive
|
|
|
|
value means unknown, at which point the number
|
|
|
|
of columns will be determined by the size of
|
|
|
|
the first row.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
"""
|
|
|
|
def __init__(self, rows, numRows=0, numCols=0):
|
|
|
|
"""
|
|
|
|
Note: This docstring is not shown publicly.
|
|
|
|
|
|
|
|
Create a wrapper over a Java RowMatrix.
|
|
|
|
|
2019-07-09 17:39:21 -04:00
|
|
|
Publicly, we require that `rows` be an RDD or DataFrame. However, for
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
internal usage, `rows` can also be a Java RowMatrix
|
|
|
|
object, in which case we can wrap it directly. This
|
|
|
|
assists in clean matrix conversions.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2, 3], [4, 5, 6]])
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> mat_diff = RowMatrix(rows)
|
|
|
|
>>> (mat_diff._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
False
|
|
|
|
|
|
|
|
>>> mat_same = RowMatrix(mat._java_matrix_wrapper._java_model)
|
|
|
|
>>> (mat_same._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
True
|
|
|
|
"""
|
|
|
|
if isinstance(rows, RDD):
|
|
|
|
rows = rows.map(_convert_to_vector)
|
2020-07-13 22:22:44 -04:00
|
|
|
java_matrix = callMLlibFunc("createRowMatrix", rows, int(numRows), int(numCols))
|
2019-07-09 17:39:21 -04:00
|
|
|
elif isinstance(rows, DataFrame):
|
2020-07-13 22:22:44 -04:00
|
|
|
java_matrix = callMLlibFunc("createRowMatrix", rows, int(numRows), int(numCols))
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
elif (isinstance(rows, JavaObject)
|
|
|
|
and rows.getClass().getSimpleName() == "RowMatrix"):
|
|
|
|
java_matrix = rows
|
|
|
|
else:
|
|
|
|
raise TypeError("rows should be an RDD of vectors, got %s" % type(rows))
|
|
|
|
|
|
|
|
self._java_matrix_wrapper = JavaModelWrapper(java_matrix)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def rows(self):
|
|
|
|
"""
|
|
|
|
Rows of the RowMatrix stored as an RDD of vectors.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> mat = RowMatrix(sc.parallelize([[1, 2, 3], [4, 5, 6]]))
|
|
|
|
>>> rows = mat.rows
|
|
|
|
>>> rows.first()
|
|
|
|
DenseVector([1.0, 2.0, 3.0])
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("rows")
|
|
|
|
|
|
|
|
def numRows(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of rows.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2, 3], [4, 5, 6],
|
|
|
|
... [7, 8, 9], [10, 11, 12]])
|
|
|
|
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
4
|
|
|
|
|
|
|
|
>>> mat = RowMatrix(rows, 7, 6)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numRows")
|
|
|
|
|
|
|
|
def numCols(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of cols.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2, 3], [4, 5, 6],
|
|
|
|
... [7, 8, 9], [10, 11, 12]])
|
|
|
|
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
3
|
|
|
|
|
|
|
|
>>> mat = RowMatrix(rows, 7, 6)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
6
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numCols")
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
def computeColumnSummaryStatistics(self):
|
|
|
|
"""
|
|
|
|
Computes column-wise summary statistics.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`MultivariateStatisticalSummary`
|
|
|
|
object containing column-wise summary statistics.
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2, 3], [4, 5, 6]])
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> colStats = mat.computeColumnSummaryStatistics()
|
|
|
|
>>> colStats.mean()
|
|
|
|
array([ 2.5, 3.5, 4.5])
|
|
|
|
"""
|
|
|
|
java_col_stats = self._java_matrix_wrapper.call("computeColumnSummaryStatistics")
|
|
|
|
return MultivariateStatisticalSummary(java_col_stats)
|
|
|
|
|
|
|
|
def computeCovariance(self):
|
|
|
|
"""
|
|
|
|
Computes the covariance matrix, treating each row as an
|
2016-11-22 06:40:18 -05:00
|
|
|
observation.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Notes
|
|
|
|
-----
|
|
|
|
This cannot be computed on matrices with more than 65535 columns.
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2], [2, 1]])
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> mat.computeCovariance()
|
|
|
|
DenseMatrix(2, 2, [0.5, -0.5, -0.5, 0.5], 0)
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("computeCovariance")
|
|
|
|
|
|
|
|
def computeGramianMatrix(self):
|
|
|
|
"""
|
2016-11-22 06:40:18 -05:00
|
|
|
Computes the Gramian matrix `A^T A`.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Notes
|
|
|
|
-----
|
|
|
|
This cannot be computed on matrices with more than 65535 columns.
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2, 3], [4, 5, 6]])
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> mat.computeGramianMatrix()
|
|
|
|
DenseMatrix(3, 3, [17.0, 22.0, 27.0, 22.0, 29.0, 36.0, 27.0, 36.0, 45.0], 0)
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("computeGramianMatrix")
|
|
|
|
|
|
|
|
@since('2.0.0')
|
|
|
|
def columnSimilarities(self, threshold=0.0):
|
|
|
|
"""
|
|
|
|
Compute similarities between columns of this matrix.
|
|
|
|
|
|
|
|
The threshold parameter is a trade-off knob between estimate
|
|
|
|
quality and computational cost.
|
|
|
|
|
|
|
|
The default threshold setting of 0 guarantees deterministically
|
|
|
|
correct results, but uses the brute-force approach of computing
|
|
|
|
normalized dot products.
|
|
|
|
|
|
|
|
Setting the threshold to positive values uses a sampling
|
|
|
|
approach and incurs strictly less computational cost than the
|
|
|
|
brute-force approach. However the similarities computed will
|
|
|
|
be estimates.
|
|
|
|
|
|
|
|
The sampling guarantees relative-error correctness for those
|
|
|
|
pairs of columns that have similarity greater than the given
|
|
|
|
similarity threshold.
|
|
|
|
|
|
|
|
To describe the guarantee, we set some notation:
|
2020-11-24 20:24:41 -05:00
|
|
|
|
|
|
|
- Let A be the smallest in magnitude non-zero element of
|
|
|
|
this matrix.
|
|
|
|
- Let B be the largest in magnitude non-zero element of
|
|
|
|
this matrix.
|
|
|
|
- Let L be the maximum number of non-zeros per row.
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
|
|
|
For example, for {0,1} matrices: A=B=1.
|
|
|
|
Another example, for the Netflix matrix: A=1, B=5
|
|
|
|
|
|
|
|
For those column pairs that are above the threshold, the
|
|
|
|
computed similarity is correct to within 20% relative error
|
|
|
|
with probability at least 1 - (0.981)^10/B^
|
|
|
|
|
|
|
|
The shuffle size is bounded by the *smaller* of the following
|
|
|
|
two expressions:
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
- O(n log(n) L / (threshold * A))
|
|
|
|
- O(m L^2^)
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
|
|
|
The latter is the cost of the brute-force approach, so for
|
|
|
|
non-zero thresholds, the cost is always cheaper than the
|
|
|
|
brute-force approach.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
threshold : float, optional
|
|
|
|
Set to 0 for deterministic guaranteed
|
|
|
|
correctness. Similarities above this
|
|
|
|
threshold are estimated with the cost vs
|
|
|
|
estimate quality trade-off described above.
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`CoordinateMatrix`
|
|
|
|
An n x n sparse upper-triangular CoordinateMatrix of
|
|
|
|
cosine similarities between columns of this matrix.
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2], [1, 5]])
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> sims = mat.columnSimilarities()
|
|
|
|
>>> sims.entries.first().value
|
|
|
|
0.91914503...
|
|
|
|
"""
|
|
|
|
java_sims_mat = self._java_matrix_wrapper.call("columnSimilarities", float(threshold))
|
|
|
|
return CoordinateMatrix(java_sims_mat)
|
|
|
|
|
|
|
|
def tallSkinnyQR(self, computeQ=False):
|
|
|
|
"""
|
|
|
|
Compute the QR decomposition of this RowMatrix.
|
|
|
|
|
|
|
|
The implementation is designed to optimize the QR decomposition
|
2020-11-24 20:24:41 -05:00
|
|
|
(factorization) for the RowMatrix of a tall and skinny shape [1]_.
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. [1] Paul G. Constantine, David F. Gleich. "Tall and skinny QR
|
|
|
|
factorizations in MapReduce architectures"
|
|
|
|
https://doi.org/10.1145/1996092.1996103
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
computeQ : bool, optional
|
|
|
|
whether to computeQ
|
|
|
|
|
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`pyspark.mllib.linalg.QRDecomposition`
|
|
|
|
QRDecomposition(Q: RowMatrix, R: Matrix), where
|
|
|
|
Q = None if computeQ = false.
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> rows = sc.parallelize([[3, -6], [4, -8], [0, 1]])
|
|
|
|
>>> mat = RowMatrix(rows)
|
|
|
|
>>> decomp = mat.tallSkinnyQR(True)
|
|
|
|
>>> Q = decomp.Q
|
|
|
|
>>> R = decomp.R
|
|
|
|
|
|
|
|
>>> # Test with absolute values
|
|
|
|
>>> absQRows = Q.rows.map(lambda row: abs(row.toArray()).tolist())
|
|
|
|
>>> absQRows.collect()
|
|
|
|
[[0.6..., 0.0], [0.8..., 0.0], [0.0, 1.0]]
|
|
|
|
|
|
|
|
>>> # Test with absolute values
|
|
|
|
>>> abs(R.toArray()).tolist()
|
|
|
|
[[5.0, 10.0], [0.0, 1.0]]
|
|
|
|
"""
|
|
|
|
decomp = JavaModelWrapper(self._java_matrix_wrapper.call("tallSkinnyQR", computeQ))
|
|
|
|
if computeQ:
|
|
|
|
java_Q = decomp.call("Q")
|
|
|
|
Q = RowMatrix(java_Q)
|
|
|
|
else:
|
|
|
|
Q = None
|
|
|
|
R = decomp.call("R")
|
|
|
|
return QRDecomposition(Q, R)
|
|
|
|
|
2017-05-03 04:58:05 -04:00
|
|
|
def computeSVD(self, k, computeU=False, rCond=1e-9):
|
|
|
|
"""
|
|
|
|
Computes the singular value decomposition of the RowMatrix.
|
|
|
|
|
|
|
|
The given row matrix A of dimension (m X n) is decomposed into
|
|
|
|
U * s * V'T where
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
- U: (m X k) (left singular vectors) is a RowMatrix whose
|
|
|
|
columns are the eigenvectors of (A X A')
|
|
|
|
- s: DenseVector consisting of square root of the eigenvalues
|
|
|
|
(singular values) in descending order.
|
|
|
|
- v: (n X k) (right singular vectors) is a Matrix whose columns
|
|
|
|
are the eigenvectors of (A' X A)
|
2017-05-03 04:58:05 -04:00
|
|
|
|
|
|
|
For more specific details on implementation, please refer
|
|
|
|
the Scala documentation.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.2.0
|
|
|
|
|
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
k : int
|
|
|
|
Number of leading singular values to keep (`0 < k <= n`).
|
|
|
|
It might return less than k if there are numerically zero singular values
|
|
|
|
or there are not enough Ritz values converged before the maximum number of
|
|
|
|
Arnoldi update iterations is reached (in case that matrix A is ill-conditioned).
|
|
|
|
computeU : bool, optional
|
|
|
|
Whether or not to compute U. If set to be
|
|
|
|
True, then U is computed by A * V * s^-1
|
|
|
|
rCond : float, optional
|
|
|
|
Reciprocal condition number. All singular values
|
|
|
|
smaller than rCond * s[0] are treated as zero
|
|
|
|
where s[0] is the largest singular value.
|
|
|
|
|
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`SingularValueDecomposition`
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2017-05-03 04:58:05 -04:00
|
|
|
>>> rows = sc.parallelize([[3, 1, 1], [-1, 3, 1]])
|
|
|
|
>>> rm = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> svd_model = rm.computeSVD(2, True)
|
|
|
|
>>> svd_model.U.rows.collect()
|
|
|
|
[DenseVector([-0.7071, 0.7071]), DenseVector([-0.7071, -0.7071])]
|
|
|
|
>>> svd_model.s
|
|
|
|
DenseVector([3.4641, 3.1623])
|
|
|
|
>>> svd_model.V
|
|
|
|
DenseMatrix(3, 2, [-0.4082, -0.8165, -0.4082, 0.8944, -0.4472, 0.0], 0)
|
|
|
|
"""
|
|
|
|
j_model = self._java_matrix_wrapper.call(
|
|
|
|
"computeSVD", int(k), bool(computeU), float(rCond))
|
|
|
|
return SingularValueDecomposition(j_model)
|
|
|
|
|
|
|
|
def computePrincipalComponents(self, k):
|
|
|
|
"""
|
|
|
|
Computes the k principal components of the given row matrix
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.2.0
|
|
|
|
|
|
|
|
Notes
|
|
|
|
-----
|
|
|
|
This cannot be computed on matrices with more than 65535 columns.
|
2017-05-03 04:58:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
k : int
|
|
|
|
Number of principal components to keep.
|
2017-05-03 04:58:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`pyspark.mllib.linalg.DenseMatrix`
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2017-05-03 04:58:05 -04:00
|
|
|
>>> rows = sc.parallelize([[1, 2, 3], [2, 4, 5], [3, 6, 1]])
|
|
|
|
>>> rm = RowMatrix(rows)
|
|
|
|
|
|
|
|
>>> # Returns the two principal components of rm
|
|
|
|
>>> pca = rm.computePrincipalComponents(2)
|
|
|
|
>>> pca
|
|
|
|
DenseMatrix(3, 2, [-0.349, -0.6981, 0.6252, -0.2796, -0.5592, -0.7805], 0)
|
|
|
|
|
|
|
|
>>> # Transform into new dimensions with the greatest variance.
|
|
|
|
>>> rm.multiply(pca).rows.collect() # doctest: +NORMALIZE_WHITESPACE
|
|
|
|
[DenseVector([0.1305, -3.7394]), DenseVector([-0.3642, -6.6983]), \
|
|
|
|
DenseVector([-4.6102, -4.9745])]
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("computePrincipalComponents", k)
|
|
|
|
|
|
|
|
def multiply(self, matrix):
|
|
|
|
"""
|
|
|
|
Multiply this matrix by a local dense matrix on the right.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.2.0
|
|
|
|
|
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
matrix : :py:class:`pyspark.mllib.linalg.Matrix`
|
|
|
|
a local dense matrix whose number of rows must match the number of columns
|
|
|
|
of this matrix
|
2017-05-03 04:58:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`RowMatrix`
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2017-05-03 04:58:05 -04:00
|
|
|
>>> rm = RowMatrix(sc.parallelize([[0, 1], [2, 3]]))
|
|
|
|
>>> rm.multiply(DenseMatrix(2, 2, [0, 2, 1, 3])).rows.collect()
|
|
|
|
[DenseVector([2.0, 3.0]), DenseVector([6.0, 11.0])]
|
|
|
|
"""
|
|
|
|
if not isinstance(matrix, DenseMatrix):
|
|
|
|
raise ValueError("Only multiplication with DenseMatrix "
|
|
|
|
"is supported.")
|
|
|
|
j_model = self._java_matrix_wrapper.call("multiply", matrix)
|
|
|
|
return RowMatrix(j_model)
|
|
|
|
|
|
|
|
|
|
|
|
class SingularValueDecomposition(JavaModelWrapper):
|
|
|
|
"""
|
|
|
|
Represents singular value decomposition (SVD) factors.
|
|
|
|
|
|
|
|
.. versionadded:: 2.2.0
|
|
|
|
"""
|
|
|
|
|
|
|
|
@property
|
|
|
|
@since('2.2.0')
|
|
|
|
def U(self):
|
|
|
|
"""
|
|
|
|
Returns a distributed matrix whose columns are the left
|
|
|
|
singular vectors of the SingularValueDecomposition if computeU was set to be True.
|
|
|
|
"""
|
|
|
|
u = self.call("U")
|
|
|
|
if u is not None:
|
|
|
|
mat_name = u.getClass().getSimpleName()
|
|
|
|
if mat_name == "RowMatrix":
|
|
|
|
return RowMatrix(u)
|
|
|
|
elif mat_name == "IndexedRowMatrix":
|
|
|
|
return IndexedRowMatrix(u)
|
|
|
|
else:
|
|
|
|
raise TypeError("Expected RowMatrix/IndexedRowMatrix got %s" % mat_name)
|
|
|
|
|
|
|
|
@property
|
|
|
|
@since('2.2.0')
|
|
|
|
def s(self):
|
|
|
|
"""
|
|
|
|
Returns a DenseVector with singular values in descending order.
|
|
|
|
"""
|
|
|
|
return self.call("s")
|
|
|
|
|
|
|
|
@property
|
|
|
|
@since('2.2.0')
|
|
|
|
def V(self):
|
|
|
|
"""
|
|
|
|
Returns a DenseMatrix whose columns are the right singular
|
|
|
|
vectors of the SingularValueDecomposition.
|
|
|
|
"""
|
|
|
|
return self.call("V")
|
|
|
|
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
class IndexedRow(object):
|
|
|
|
"""
|
|
|
|
Represents a row of an IndexedRowMatrix.
|
|
|
|
|
2020-07-13 22:22:44 -04:00
|
|
|
Just a wrapper over a (int, vector) tuple.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
index : int
|
|
|
|
The index for the given row.
|
|
|
|
vector : :py:class:`pyspark.mllib.linalg.Vector` or convertible
|
|
|
|
The row in the matrix at the given index.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
"""
|
|
|
|
def __init__(self, index, vector):
|
2020-07-13 22:22:44 -04:00
|
|
|
self.index = int(index)
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
self.vector = _convert_to_vector(vector)
|
|
|
|
|
|
|
|
def __repr__(self):
|
|
|
|
return "IndexedRow(%s, %s)" % (self.index, self.vector)
|
|
|
|
|
|
|
|
|
|
|
|
def _convert_to_indexed_row(row):
|
|
|
|
if isinstance(row, IndexedRow):
|
|
|
|
return row
|
|
|
|
elif isinstance(row, tuple) and len(row) == 2:
|
|
|
|
return IndexedRow(*row)
|
|
|
|
else:
|
|
|
|
raise TypeError("Cannot convert type %s into IndexedRow" % type(row))
|
|
|
|
|
|
|
|
|
|
|
|
class IndexedRowMatrix(DistributedMatrix):
|
|
|
|
"""
|
|
|
|
Represents a row-oriented distributed Matrix with indexed rows.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
rows : :py:class:`pyspark.RDD`
|
|
|
|
An RDD of IndexedRows or (int, vector) tuples or a DataFrame consisting of a
|
|
|
|
int typed column of indices and a vector typed column.
|
|
|
|
numRows : int, optional
|
|
|
|
Number of rows in the matrix. A non-positive
|
|
|
|
value means unknown, at which point the number
|
|
|
|
of rows will be determined by the max row
|
|
|
|
index plus one.
|
|
|
|
numCols : int, optional
|
|
|
|
Number of columns in the matrix. A non-positive
|
|
|
|
value means unknown, at which point the number
|
|
|
|
of columns will be determined by the size of
|
|
|
|
the first row.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
"""
|
|
|
|
def __init__(self, rows, numRows=0, numCols=0):
|
|
|
|
"""
|
|
|
|
Note: This docstring is not shown publicly.
|
|
|
|
|
|
|
|
Create a wrapper over a Java IndexedRowMatrix.
|
|
|
|
|
2019-07-09 17:39:21 -04:00
|
|
|
Publicly, we require that `rows` be an RDD or DataFrame. However, for
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
internal usage, `rows` can also be a Java IndexedRowMatrix
|
|
|
|
object, in which case we can wrap it directly. This
|
|
|
|
assists in clean matrix conversions.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(1, [4, 5, 6])])
|
|
|
|
>>> mat = IndexedRowMatrix(rows)
|
|
|
|
|
|
|
|
>>> mat_diff = IndexedRowMatrix(rows)
|
|
|
|
>>> (mat_diff._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
False
|
|
|
|
|
|
|
|
>>> mat_same = IndexedRowMatrix(mat._java_matrix_wrapper._java_model)
|
|
|
|
>>> (mat_same._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
True
|
|
|
|
"""
|
|
|
|
if isinstance(rows, RDD):
|
|
|
|
rows = rows.map(_convert_to_indexed_row)
|
|
|
|
# We use DataFrames for serialization of IndexedRows from
|
|
|
|
# Python, so first convert the RDD to a DataFrame on this
|
|
|
|
# side. This will convert each IndexedRow to a Row
|
|
|
|
# containing the 'index' and 'vector' values, which can
|
|
|
|
# both be easily serialized. We will convert back to
|
|
|
|
# IndexedRows on the Scala side.
|
|
|
|
java_matrix = callMLlibFunc("createIndexedRowMatrix", rows.toDF(),
|
2020-07-13 22:22:44 -04:00
|
|
|
int(numRows), int(numCols))
|
2019-07-09 17:39:21 -04:00
|
|
|
elif isinstance(rows, DataFrame):
|
2020-07-13 22:22:44 -04:00
|
|
|
java_matrix = callMLlibFunc("createIndexedRowMatrix", rows, int(numRows), int(numCols))
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
elif (isinstance(rows, JavaObject)
|
|
|
|
and rows.getClass().getSimpleName() == "IndexedRowMatrix"):
|
|
|
|
java_matrix = rows
|
|
|
|
else:
|
2020-07-13 22:22:44 -04:00
|
|
|
raise TypeError("rows should be an RDD of IndexedRows or (int, vector) tuples, "
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
"got %s" % type(rows))
|
|
|
|
|
|
|
|
self._java_matrix_wrapper = JavaModelWrapper(java_matrix)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def rows(self):
|
|
|
|
"""
|
|
|
|
Rows of the IndexedRowMatrix stored as an RDD of IndexedRows.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> mat = IndexedRowMatrix(sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(1, [4, 5, 6])]))
|
|
|
|
>>> rows = mat.rows
|
|
|
|
>>> rows.first()
|
|
|
|
IndexedRow(0, [1.0,2.0,3.0])
|
|
|
|
"""
|
|
|
|
# We use DataFrames for serialization of IndexedRows from
|
|
|
|
# Java, so we first convert the RDD of rows to a DataFrame
|
|
|
|
# on the Scala/Java side. Then we map each Row in the
|
|
|
|
# DataFrame back to an IndexedRow on this side.
|
|
|
|
rows_df = callMLlibFunc("getIndexedRows", self._java_matrix_wrapper._java_model)
|
2016-03-02 18:26:34 -05:00
|
|
|
rows = rows_df.rdd.map(lambda row: IndexedRow(row[0], row[1]))
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
return rows
|
|
|
|
|
|
|
|
def numRows(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of rows.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(1, [4, 5, 6]),
|
|
|
|
... IndexedRow(2, [7, 8, 9]),
|
|
|
|
... IndexedRow(3, [10, 11, 12])])
|
|
|
|
|
|
|
|
>>> mat = IndexedRowMatrix(rows)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
4
|
|
|
|
|
|
|
|
>>> mat = IndexedRowMatrix(rows, 7, 6)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numRows")
|
|
|
|
|
|
|
|
def numCols(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of cols.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(1, [4, 5, 6]),
|
|
|
|
... IndexedRow(2, [7, 8, 9]),
|
|
|
|
... IndexedRow(3, [10, 11, 12])])
|
|
|
|
|
|
|
|
>>> mat = IndexedRowMatrix(rows)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
3
|
|
|
|
|
|
|
|
>>> mat = IndexedRowMatrix(rows, 7, 6)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
6
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numCols")
|
|
|
|
|
2016-01-05 18:33:27 -05:00
|
|
|
def columnSimilarities(self):
|
|
|
|
"""
|
|
|
|
Compute all cosine similarities between columns.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2016-01-05 18:33:27 -05:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(6, [4, 5, 6])])
|
|
|
|
>>> mat = IndexedRowMatrix(rows)
|
|
|
|
>>> cs = mat.columnSimilarities()
|
|
|
|
>>> print(cs.numCols())
|
|
|
|
3
|
|
|
|
"""
|
|
|
|
java_coordinate_matrix = self._java_matrix_wrapper.call("columnSimilarities")
|
|
|
|
return CoordinateMatrix(java_coordinate_matrix)
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
def computeGramianMatrix(self):
|
|
|
|
"""
|
2016-11-22 06:40:18 -05:00
|
|
|
Computes the Gramian matrix `A^T A`.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Notes
|
|
|
|
-----
|
|
|
|
This cannot be computed on matrices with more than 65535 columns.
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(1, [4, 5, 6])])
|
|
|
|
>>> mat = IndexedRowMatrix(rows)
|
|
|
|
|
|
|
|
>>> mat.computeGramianMatrix()
|
|
|
|
DenseMatrix(3, 3, [17.0, 22.0, 27.0, 22.0, 29.0, 36.0, 27.0, 36.0, 45.0], 0)
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("computeGramianMatrix")
|
|
|
|
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
def toRowMatrix(self):
|
|
|
|
"""
|
|
|
|
Convert this matrix to a RowMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(6, [4, 5, 6])])
|
|
|
|
>>> mat = IndexedRowMatrix(rows).toRowMatrix()
|
|
|
|
>>> mat.rows.collect()
|
|
|
|
[DenseVector([1.0, 2.0, 3.0]), DenseVector([4.0, 5.0, 6.0])]
|
|
|
|
"""
|
|
|
|
java_row_matrix = self._java_matrix_wrapper.call("toRowMatrix")
|
|
|
|
return RowMatrix(java_row_matrix)
|
|
|
|
|
|
|
|
def toCoordinateMatrix(self):
|
|
|
|
"""
|
|
|
|
Convert this matrix to a CoordinateMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 0]),
|
|
|
|
... IndexedRow(6, [0, 5])])
|
|
|
|
>>> mat = IndexedRowMatrix(rows).toCoordinateMatrix()
|
|
|
|
>>> mat.entries.take(3)
|
|
|
|
[MatrixEntry(0, 0, 1.0), MatrixEntry(0, 1, 0.0), MatrixEntry(6, 0, 0.0)]
|
|
|
|
"""
|
|
|
|
java_coordinate_matrix = self._java_matrix_wrapper.call("toCoordinateMatrix")
|
|
|
|
return CoordinateMatrix(java_coordinate_matrix)
|
|
|
|
|
2015-08-05 10:40:50 -04:00
|
|
|
def toBlockMatrix(self, rowsPerBlock=1024, colsPerBlock=1024):
|
|
|
|
"""
|
|
|
|
Convert this matrix to a BlockMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
rowsPerBlock : int, optional
|
|
|
|
Number of rows that make up each block.
|
|
|
|
The blocks forming the final rows are not
|
|
|
|
required to have the given number of rows.
|
|
|
|
colsPerBlock : int, optional
|
|
|
|
Number of columns that make up each block.
|
|
|
|
The blocks forming the final columns are not
|
|
|
|
required to have the given number of columns.
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> rows = sc.parallelize([IndexedRow(0, [1, 2, 3]),
|
|
|
|
... IndexedRow(6, [4, 5, 6])])
|
|
|
|
>>> mat = IndexedRowMatrix(rows).toBlockMatrix()
|
|
|
|
|
|
|
|
>>> # This IndexedRowMatrix will have 7 effective rows, due to
|
|
|
|
>>> # the highest row index being 6, and the ensuing
|
|
|
|
>>> # BlockMatrix will have 7 rows as well.
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
3
|
|
|
|
"""
|
|
|
|
java_block_matrix = self._java_matrix_wrapper.call("toBlockMatrix",
|
|
|
|
rowsPerBlock,
|
|
|
|
colsPerBlock)
|
|
|
|
return BlockMatrix(java_block_matrix, rowsPerBlock, colsPerBlock)
|
|
|
|
|
2017-05-03 04:58:05 -04:00
|
|
|
def computeSVD(self, k, computeU=False, rCond=1e-9):
|
|
|
|
"""
|
|
|
|
Computes the singular value decomposition of the IndexedRowMatrix.
|
|
|
|
|
|
|
|
The given row matrix A of dimension (m X n) is decomposed into
|
|
|
|
U * s * V'T where
|
|
|
|
|
|
|
|
* U: (m X k) (left singular vectors) is a IndexedRowMatrix
|
|
|
|
whose columns are the eigenvectors of (A X A')
|
|
|
|
* s: DenseVector consisting of square root of the eigenvalues
|
|
|
|
(singular values) in descending order.
|
|
|
|
* v: (n X k) (right singular vectors) is a Matrix whose columns
|
|
|
|
are the eigenvectors of (A' X A)
|
|
|
|
|
|
|
|
For more specific details on implementation, please refer
|
|
|
|
the scala documentation.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.2.0
|
|
|
|
|
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
k : int
|
|
|
|
Number of leading singular values to keep (`0 < k <= n`).
|
|
|
|
It might return less than k if there are numerically zero singular values
|
|
|
|
or there are not enough Ritz values converged before the maximum number of
|
|
|
|
Arnoldi update iterations is reached (in case that matrix A is ill-conditioned).
|
|
|
|
computeU : bool, optional
|
|
|
|
Whether or not to compute U. If set to be
|
|
|
|
True, then U is computed by A * V * s^-1
|
|
|
|
rCond : float, optional
|
|
|
|
Reciprocal condition number. All singular values
|
|
|
|
smaller than rCond * s[0] are treated as zero
|
|
|
|
where s[0] is the largest singular value.
|
|
|
|
|
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`SingularValueDecomposition`
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2017-05-03 04:58:05 -04:00
|
|
|
>>> rows = [(0, (3, 1, 1)), (1, (-1, 3, 1))]
|
|
|
|
>>> irm = IndexedRowMatrix(sc.parallelize(rows))
|
|
|
|
>>> svd_model = irm.computeSVD(2, True)
|
|
|
|
>>> svd_model.U.rows.collect() # doctest: +NORMALIZE_WHITESPACE
|
|
|
|
[IndexedRow(0, [-0.707106781187,0.707106781187]),\
|
|
|
|
IndexedRow(1, [-0.707106781187,-0.707106781187])]
|
|
|
|
>>> svd_model.s
|
|
|
|
DenseVector([3.4641, 3.1623])
|
|
|
|
>>> svd_model.V
|
|
|
|
DenseMatrix(3, 2, [-0.4082, -0.8165, -0.4082, 0.8944, -0.4472, 0.0], 0)
|
|
|
|
"""
|
|
|
|
j_model = self._java_matrix_wrapper.call(
|
|
|
|
"computeSVD", int(k), bool(computeU), float(rCond))
|
|
|
|
return SingularValueDecomposition(j_model)
|
|
|
|
|
|
|
|
def multiply(self, matrix):
|
|
|
|
"""
|
|
|
|
Multiply this matrix by a local dense matrix on the right.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.2.0
|
|
|
|
|
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
matrix : :py:class:`pyspark.mllib.linalg.Matrix`
|
|
|
|
a local dense matrix whose number of rows must match the number of columns
|
|
|
|
of this matrix
|
2017-05-03 04:58:05 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Returns
|
|
|
|
-------
|
|
|
|
:py:class:`IndexedRowMatrix`
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2017-05-03 04:58:05 -04:00
|
|
|
>>> mat = IndexedRowMatrix(sc.parallelize([(0, (0, 1)), (1, (2, 3))]))
|
|
|
|
>>> mat.multiply(DenseMatrix(2, 2, [0, 2, 1, 3])).rows.collect()
|
|
|
|
[IndexedRow(0, [2.0,3.0]), IndexedRow(1, [6.0,11.0])]
|
|
|
|
"""
|
|
|
|
if not isinstance(matrix, DenseMatrix):
|
|
|
|
raise ValueError("Only multiplication with DenseMatrix "
|
|
|
|
"is supported.")
|
|
|
|
return IndexedRowMatrix(self._java_matrix_wrapper.call("multiply", matrix))
|
|
|
|
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
class MatrixEntry(object):
|
|
|
|
"""
|
|
|
|
Represents an entry of a CoordinateMatrix.
|
|
|
|
|
2020-07-13 22:22:44 -04:00
|
|
|
Just a wrapper over a (int, int, float) tuple.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
i : int
|
|
|
|
The row index of the matrix.
|
|
|
|
j : int
|
|
|
|
The column index of the matrix.
|
|
|
|
value : float
|
|
|
|
The (i, j)th entry of the matrix, as a float.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
"""
|
|
|
|
def __init__(self, i, j, value):
|
2020-07-13 22:22:44 -04:00
|
|
|
self.i = int(i)
|
|
|
|
self.j = int(j)
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
self.value = float(value)
|
|
|
|
|
|
|
|
def __repr__(self):
|
|
|
|
return "MatrixEntry(%s, %s, %s)" % (self.i, self.j, self.value)
|
|
|
|
|
|
|
|
|
|
|
|
def _convert_to_matrix_entry(entry):
|
|
|
|
if isinstance(entry, MatrixEntry):
|
|
|
|
return entry
|
|
|
|
elif isinstance(entry, tuple) and len(entry) == 3:
|
|
|
|
return MatrixEntry(*entry)
|
|
|
|
else:
|
|
|
|
raise TypeError("Cannot convert type %s into MatrixEntry" % type(entry))
|
|
|
|
|
|
|
|
|
|
|
|
class CoordinateMatrix(DistributedMatrix):
|
|
|
|
"""
|
|
|
|
Represents a matrix in coordinate format.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
entries : :py:class:`pyspark.RDD`
|
|
|
|
An RDD of MatrixEntry inputs or
|
|
|
|
(int, int, float) tuples.
|
|
|
|
numRows : int, optional
|
|
|
|
Number of rows in the matrix. A non-positive
|
|
|
|
value means unknown, at which point the number
|
|
|
|
of rows will be determined by the max row
|
|
|
|
index plus one.
|
|
|
|
numCols : int, optional
|
|
|
|
Number of columns in the matrix. A non-positive
|
|
|
|
value means unknown, at which point the number
|
|
|
|
of columns will be determined by the max row
|
|
|
|
index plus one.
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
"""
|
|
|
|
def __init__(self, entries, numRows=0, numCols=0):
|
|
|
|
"""
|
|
|
|
Note: This docstring is not shown publicly.
|
|
|
|
|
|
|
|
Create a wrapper over a Java CoordinateMatrix.
|
|
|
|
|
|
|
|
Publicly, we require that `rows` be an RDD. However, for
|
|
|
|
internal usage, `rows` can also be a Java CoordinateMatrix
|
|
|
|
object, in which case we can wrap it directly. This
|
|
|
|
assists in clean matrix conversions.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(6, 4, 2.1)])
|
|
|
|
>>> mat = CoordinateMatrix(entries)
|
|
|
|
|
|
|
|
>>> mat_diff = CoordinateMatrix(entries)
|
|
|
|
>>> (mat_diff._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
False
|
|
|
|
|
|
|
|
>>> mat_same = CoordinateMatrix(mat._java_matrix_wrapper._java_model)
|
|
|
|
>>> (mat_same._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
True
|
|
|
|
"""
|
|
|
|
if isinstance(entries, RDD):
|
|
|
|
entries = entries.map(_convert_to_matrix_entry)
|
|
|
|
# We use DataFrames for serialization of MatrixEntry entries
|
|
|
|
# from Python, so first convert the RDD to a DataFrame on
|
|
|
|
# this side. This will convert each MatrixEntry to a Row
|
|
|
|
# containing the 'i', 'j', and 'value' values, which can
|
|
|
|
# each be easily serialized. We will convert back to
|
|
|
|
# MatrixEntry inputs on the Scala side.
|
|
|
|
java_matrix = callMLlibFunc("createCoordinateMatrix", entries.toDF(),
|
2020-07-13 22:22:44 -04:00
|
|
|
int(numRows), int(numCols))
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
elif (isinstance(entries, JavaObject)
|
|
|
|
and entries.getClass().getSimpleName() == "CoordinateMatrix"):
|
|
|
|
java_matrix = entries
|
|
|
|
else:
|
|
|
|
raise TypeError("entries should be an RDD of MatrixEntry entries or "
|
2020-07-13 22:22:44 -04:00
|
|
|
"(int, int, float) tuples, got %s" % type(entries))
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
self._java_matrix_wrapper = JavaModelWrapper(java_matrix)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def entries(self):
|
|
|
|
"""
|
|
|
|
Entries of the CoordinateMatrix stored as an RDD of
|
|
|
|
MatrixEntries.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> mat = CoordinateMatrix(sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(6, 4, 2.1)]))
|
|
|
|
>>> entries = mat.entries
|
|
|
|
>>> entries.first()
|
|
|
|
MatrixEntry(0, 0, 1.2)
|
|
|
|
"""
|
|
|
|
# We use DataFrames for serialization of MatrixEntry entries
|
|
|
|
# from Java, so we first convert the RDD of entries to a
|
|
|
|
# DataFrame on the Scala/Java side. Then we map each Row in
|
|
|
|
# the DataFrame back to a MatrixEntry on this side.
|
|
|
|
entries_df = callMLlibFunc("getMatrixEntries", self._java_matrix_wrapper._java_model)
|
2016-03-02 18:26:34 -05:00
|
|
|
entries = entries_df.rdd.map(lambda row: MatrixEntry(row[0], row[1], row[2]))
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
return entries
|
|
|
|
|
|
|
|
def numRows(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of rows.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(1, 0, 2),
|
|
|
|
... MatrixEntry(2, 1, 3.7)])
|
|
|
|
|
|
|
|
>>> mat = CoordinateMatrix(entries)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
3
|
|
|
|
|
|
|
|
>>> mat = CoordinateMatrix(entries, 7, 6)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numRows")
|
|
|
|
|
|
|
|
def numCols(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of cols.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(1, 0, 2),
|
|
|
|
... MatrixEntry(2, 1, 3.7)])
|
|
|
|
|
|
|
|
>>> mat = CoordinateMatrix(entries)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
2
|
|
|
|
|
|
|
|
>>> mat = CoordinateMatrix(entries, 7, 6)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
6
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numCols")
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
def transpose(self):
|
|
|
|
"""
|
|
|
|
Transpose this CoordinateMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(1, 0, 2),
|
|
|
|
... MatrixEntry(2, 1, 3.7)])
|
|
|
|
>>> mat = CoordinateMatrix(entries)
|
|
|
|
>>> mat_transposed = mat.transpose()
|
|
|
|
|
|
|
|
>>> print(mat_transposed.numRows())
|
|
|
|
2
|
|
|
|
|
|
|
|
>>> print(mat_transposed.numCols())
|
|
|
|
3
|
|
|
|
"""
|
|
|
|
java_transposed_matrix = self._java_matrix_wrapper.call("transpose")
|
|
|
|
return CoordinateMatrix(java_transposed_matrix)
|
|
|
|
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
def toRowMatrix(self):
|
|
|
|
"""
|
|
|
|
Convert this matrix to a RowMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(6, 4, 2.1)])
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> mat = CoordinateMatrix(entries).toRowMatrix()
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
>>> # This CoordinateMatrix will have 7 effective rows, due to
|
|
|
|
>>> # the highest row index being 6, but the ensuing RowMatrix
|
|
|
|
>>> # will only have 2 rows since there are only entries on 2
|
|
|
|
>>> # unique rows.
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
2
|
|
|
|
|
|
|
|
>>> # This CoordinateMatrix will have 5 columns, due to the
|
|
|
|
>>> # highest column index being 4, and the ensuing RowMatrix
|
|
|
|
>>> # will have 5 columns as well.
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
5
|
|
|
|
"""
|
|
|
|
java_row_matrix = self._java_matrix_wrapper.call("toRowMatrix")
|
|
|
|
return RowMatrix(java_row_matrix)
|
|
|
|
|
|
|
|
def toIndexedRowMatrix(self):
|
|
|
|
"""
|
|
|
|
Convert this matrix to an IndexedRowMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(6, 4, 2.1)])
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> mat = CoordinateMatrix(entries).toIndexedRowMatrix()
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
>>> # This CoordinateMatrix will have 7 effective rows, due to
|
|
|
|
>>> # the highest row index being 6, and the ensuing
|
|
|
|
>>> # IndexedRowMatrix will have 7 rows as well.
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
|
|
|
|
>>> # This CoordinateMatrix will have 5 columns, due to the
|
|
|
|
>>> # highest column index being 4, and the ensuing
|
|
|
|
>>> # IndexedRowMatrix will have 5 columns as well.
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
5
|
|
|
|
"""
|
|
|
|
java_indexed_row_matrix = self._java_matrix_wrapper.call("toIndexedRowMatrix")
|
|
|
|
return IndexedRowMatrix(java_indexed_row_matrix)
|
|
|
|
|
2015-08-05 10:40:50 -04:00
|
|
|
def toBlockMatrix(self, rowsPerBlock=1024, colsPerBlock=1024):
|
|
|
|
"""
|
|
|
|
Convert this matrix to a BlockMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
rowsPerBlock : int, optional
|
|
|
|
Number of rows that make up each block.
|
|
|
|
The blocks forming the final rows are not
|
|
|
|
required to have the given number of rows.
|
|
|
|
colsPerBlock : int, optional
|
|
|
|
Number of columns that make up each block.
|
|
|
|
The blocks forming the final columns are not
|
|
|
|
required to have the given number of columns.
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> entries = sc.parallelize([MatrixEntry(0, 0, 1.2),
|
|
|
|
... MatrixEntry(6, 4, 2.1)])
|
|
|
|
>>> mat = CoordinateMatrix(entries).toBlockMatrix()
|
|
|
|
|
|
|
|
>>> # This CoordinateMatrix will have 7 effective rows, due to
|
|
|
|
>>> # the highest row index being 6, and the ensuing
|
|
|
|
>>> # BlockMatrix will have 7 rows as well.
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
|
|
|
|
>>> # This CoordinateMatrix will have 5 columns, due to the
|
|
|
|
>>> # highest column index being 4, and the ensuing
|
|
|
|
>>> # BlockMatrix will have 5 columns as well.
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
5
|
|
|
|
"""
|
|
|
|
java_block_matrix = self._java_matrix_wrapper.call("toBlockMatrix",
|
|
|
|
rowsPerBlock,
|
|
|
|
colsPerBlock)
|
|
|
|
return BlockMatrix(java_block_matrix, rowsPerBlock, colsPerBlock)
|
|
|
|
|
|
|
|
|
|
|
|
def _convert_to_matrix_block_tuple(block):
|
|
|
|
if (isinstance(block, tuple) and len(block) == 2
|
|
|
|
and isinstance(block[0], tuple) and len(block[0]) == 2
|
|
|
|
and isinstance(block[1], Matrix)):
|
|
|
|
blockRowIndex = int(block[0][0])
|
|
|
|
blockColIndex = int(block[0][1])
|
|
|
|
subMatrix = block[1]
|
|
|
|
return ((blockRowIndex, blockColIndex), subMatrix)
|
|
|
|
else:
|
|
|
|
raise TypeError("Cannot convert type %s into a sub-matrix block tuple" % type(block))
|
|
|
|
|
|
|
|
|
|
|
|
class BlockMatrix(DistributedMatrix):
|
|
|
|
"""
|
|
|
|
Represents a distributed matrix in blocks of local matrices.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Parameters
|
|
|
|
----------
|
|
|
|
blocks : :py:class:`pyspark.RDD`
|
|
|
|
An RDD of sub-matrix blocks
|
|
|
|
((blockRowIndex, blockColIndex), sub-matrix) that
|
|
|
|
form this distributed matrix. If multiple blocks
|
|
|
|
with the same index exist, the results for
|
|
|
|
operations like add and multiply will be
|
|
|
|
unpredictable.
|
|
|
|
rowsPerBlock : int
|
|
|
|
Number of rows that make up each block.
|
|
|
|
The blocks forming the final rows are not
|
|
|
|
required to have the given number of rows.
|
|
|
|
colsPerBlock : int
|
|
|
|
Number of columns that make up each block.
|
|
|
|
The blocks forming the final columns are not
|
|
|
|
required to have the given number of columns.
|
|
|
|
numRows : int, optional
|
|
|
|
Number of rows of this matrix. If the supplied
|
|
|
|
value is less than or equal to zero, the number
|
|
|
|
of rows will be calculated when `numRows` is
|
|
|
|
invoked.
|
|
|
|
numCols : int, optional
|
|
|
|
Number of columns of this matrix. If the supplied
|
|
|
|
value is less than or equal to zero, the number
|
|
|
|
of columns will be calculated when `numCols` is
|
|
|
|
invoked.
|
2015-08-05 10:40:50 -04:00
|
|
|
"""
|
|
|
|
def __init__(self, blocks, rowsPerBlock, colsPerBlock, numRows=0, numCols=0):
|
|
|
|
"""
|
|
|
|
Note: This docstring is not shown publicly.
|
|
|
|
|
|
|
|
Create a wrapper over a Java BlockMatrix.
|
|
|
|
|
|
|
|
Publicly, we require that `blocks` be an RDD. However, for
|
|
|
|
internal usage, `blocks` can also be a Java BlockMatrix
|
|
|
|
object, in which case we can wrap it directly. This
|
|
|
|
assists in clean matrix conversions.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
|
|
|
|
>>> mat_diff = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> (mat_diff._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
False
|
|
|
|
|
|
|
|
>>> mat_same = BlockMatrix(mat._java_matrix_wrapper._java_model, 3, 2)
|
|
|
|
>>> (mat_same._java_matrix_wrapper._java_model ==
|
|
|
|
... mat._java_matrix_wrapper._java_model)
|
|
|
|
True
|
|
|
|
"""
|
|
|
|
if isinstance(blocks, RDD):
|
|
|
|
blocks = blocks.map(_convert_to_matrix_block_tuple)
|
|
|
|
# We use DataFrames for serialization of sub-matrix blocks
|
|
|
|
# from Python, so first convert the RDD to a DataFrame on
|
|
|
|
# this side. This will convert each sub-matrix block
|
|
|
|
# tuple to a Row containing the 'blockRowIndex',
|
|
|
|
# 'blockColIndex', and 'subMatrix' values, which can
|
|
|
|
# each be easily serialized. We will convert back to
|
|
|
|
# ((blockRowIndex, blockColIndex), sub-matrix) tuples on
|
|
|
|
# the Scala side.
|
|
|
|
java_matrix = callMLlibFunc("createBlockMatrix", blocks.toDF(),
|
|
|
|
int(rowsPerBlock), int(colsPerBlock),
|
2020-07-13 22:22:44 -04:00
|
|
|
int(numRows), int(numCols))
|
2015-08-05 10:40:50 -04:00
|
|
|
elif (isinstance(blocks, JavaObject)
|
|
|
|
and blocks.getClass().getSimpleName() == "BlockMatrix"):
|
|
|
|
java_matrix = blocks
|
|
|
|
else:
|
|
|
|
raise TypeError("blocks should be an RDD of sub-matrix blocks as "
|
|
|
|
"((int, int), matrix) tuples, got %s" % type(blocks))
|
|
|
|
|
|
|
|
self._java_matrix_wrapper = JavaModelWrapper(java_matrix)
|
|
|
|
|
|
|
|
@property
|
|
|
|
def blocks(self):
|
|
|
|
"""
|
|
|
|
The RDD of sub-matrix blocks
|
|
|
|
((blockRowIndex, blockColIndex), sub-matrix) that form this
|
|
|
|
distributed matrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> mat = BlockMatrix(
|
|
|
|
... sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))]), 3, 2)
|
|
|
|
>>> blocks = mat.blocks
|
|
|
|
>>> blocks.first()
|
|
|
|
((0, 0), DenseMatrix(3, 2, [1.0, 2.0, 3.0, 4.0, 5.0, 6.0], 0))
|
|
|
|
|
|
|
|
"""
|
|
|
|
# We use DataFrames for serialization of sub-matrix blocks
|
|
|
|
# from Java, so we first convert the RDD of blocks to a
|
|
|
|
# DataFrame on the Scala/Java side. Then we map each Row in
|
|
|
|
# the DataFrame back to a sub-matrix block on this side.
|
|
|
|
blocks_df = callMLlibFunc("getMatrixBlocks", self._java_matrix_wrapper._java_model)
|
2016-03-02 18:26:34 -05:00
|
|
|
blocks = blocks_df.rdd.map(lambda row: ((row[0][0], row[0][1]), row[1]))
|
2015-08-05 10:40:50 -04:00
|
|
|
return blocks
|
|
|
|
|
|
|
|
@property
|
|
|
|
def rowsPerBlock(self):
|
|
|
|
"""
|
|
|
|
Number of rows that make up each block.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> mat.rowsPerBlock
|
|
|
|
3
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("rowsPerBlock")
|
|
|
|
|
|
|
|
@property
|
|
|
|
def colsPerBlock(self):
|
|
|
|
"""
|
|
|
|
Number of columns that make up each block.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> mat.colsPerBlock
|
|
|
|
2
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("colsPerBlock")
|
|
|
|
|
|
|
|
@property
|
|
|
|
def numRowBlocks(self):
|
|
|
|
"""
|
|
|
|
Number of rows of blocks in the BlockMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> mat.numRowBlocks
|
|
|
|
2
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numRowBlocks")
|
|
|
|
|
|
|
|
@property
|
|
|
|
def numColBlocks(self):
|
|
|
|
"""
|
|
|
|
Number of columns of blocks in the BlockMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> mat.numColBlocks
|
|
|
|
1
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numColBlocks")
|
|
|
|
|
|
|
|
def numRows(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of rows.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
6
|
|
|
|
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2, 7, 6)
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
7
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numRows")
|
|
|
|
|
|
|
|
def numCols(self):
|
|
|
|
"""
|
|
|
|
Get or compute the number of cols.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
2
|
|
|
|
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2, 7, 6)
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
6
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("numCols")
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
@since('2.0.0')
|
|
|
|
def cache(self):
|
|
|
|
"""
|
|
|
|
Caches the underlying RDD.
|
|
|
|
"""
|
|
|
|
self._java_matrix_wrapper.call("cache")
|
|
|
|
return self
|
|
|
|
|
|
|
|
@since('2.0.0')
|
|
|
|
def persist(self, storageLevel):
|
|
|
|
"""
|
|
|
|
Persists the underlying RDD with the specified storage level.
|
|
|
|
"""
|
|
|
|
if not isinstance(storageLevel, StorageLevel):
|
|
|
|
raise TypeError("`storageLevel` should be a StorageLevel, got %s" % type(storageLevel))
|
|
|
|
javaStorageLevel = self._java_matrix_wrapper._sc._getJavaStorageLevel(storageLevel)
|
|
|
|
self._java_matrix_wrapper.call("persist", javaStorageLevel)
|
|
|
|
return self
|
|
|
|
|
|
|
|
@since('2.0.0')
|
|
|
|
def validate(self):
|
|
|
|
"""
|
|
|
|
Validates the block matrix info against the matrix data (`blocks`)
|
|
|
|
and throws an exception if any error is found.
|
|
|
|
"""
|
|
|
|
self._java_matrix_wrapper.call("validate")
|
|
|
|
|
2015-10-27 14:05:14 -04:00
|
|
|
def add(self, other):
|
|
|
|
"""
|
|
|
|
Adds two block matrices together. The matrices must have the
|
|
|
|
same size and matching `rowsPerBlock` and `colsPerBlock` values.
|
|
|
|
If one of the sub matrix blocks that are being added is a
|
|
|
|
SparseMatrix, the resulting sub matrix block will also be a
|
|
|
|
SparseMatrix, even if it is being added to a DenseMatrix. If
|
|
|
|
two dense sub matrix blocks are added, the output block will
|
|
|
|
also be a DenseMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-10-27 14:05:14 -04:00
|
|
|
>>> dm1 = Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])
|
|
|
|
>>> dm2 = Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12])
|
|
|
|
>>> sm = Matrices.sparse(3, 2, [0, 1, 3], [0, 1, 2], [7, 11, 12])
|
|
|
|
>>> blocks1 = sc.parallelize([((0, 0), dm1), ((1, 0), dm2)])
|
|
|
|
>>> blocks2 = sc.parallelize([((0, 0), dm1), ((1, 0), dm2)])
|
|
|
|
>>> blocks3 = sc.parallelize([((0, 0), sm), ((1, 0), dm2)])
|
|
|
|
>>> mat1 = BlockMatrix(blocks1, 3, 2)
|
|
|
|
>>> mat2 = BlockMatrix(blocks2, 3, 2)
|
|
|
|
>>> mat3 = BlockMatrix(blocks3, 3, 2)
|
|
|
|
|
|
|
|
>>> mat1.add(mat2).toLocalMatrix()
|
|
|
|
DenseMatrix(6, 2, [2.0, 4.0, 6.0, 14.0, 16.0, 18.0, 8.0, 10.0, 12.0, 20.0, 22.0, 24.0], 0)
|
|
|
|
|
|
|
|
>>> mat1.add(mat3).toLocalMatrix()
|
|
|
|
DenseMatrix(6, 2, [8.0, 2.0, 3.0, 14.0, 16.0, 18.0, 4.0, 16.0, 18.0, 20.0, 22.0, 24.0], 0)
|
|
|
|
"""
|
|
|
|
if not isinstance(other, BlockMatrix):
|
|
|
|
raise TypeError("Other should be a BlockMatrix, got %s" % type(other))
|
|
|
|
|
|
|
|
other_java_block_matrix = other._java_matrix_wrapper._java_model
|
|
|
|
java_block_matrix = self._java_matrix_wrapper.call("add", other_java_block_matrix)
|
|
|
|
return BlockMatrix(java_block_matrix, self.rowsPerBlock, self.colsPerBlock)
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
def subtract(self, other):
|
|
|
|
"""
|
|
|
|
Subtracts the given block matrix `other` from this block matrix:
|
|
|
|
`this - other`. The matrices must have the same size and
|
|
|
|
matching `rowsPerBlock` and `colsPerBlock` values. If one of
|
|
|
|
the sub matrix blocks that are being subtracted is a
|
|
|
|
SparseMatrix, the resulting sub matrix block will also be a
|
|
|
|
SparseMatrix, even if it is being subtracted from a DenseMatrix.
|
|
|
|
If two dense sub matrix blocks are subtracted, the output block
|
|
|
|
will also be a DenseMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> dm1 = Matrices.dense(3, 2, [3, 1, 5, 4, 6, 2])
|
|
|
|
>>> dm2 = Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12])
|
|
|
|
>>> sm = Matrices.sparse(3, 2, [0, 1, 3], [0, 1, 2], [1, 2, 3])
|
|
|
|
>>> blocks1 = sc.parallelize([((0, 0), dm1), ((1, 0), dm2)])
|
|
|
|
>>> blocks2 = sc.parallelize([((0, 0), dm2), ((1, 0), dm1)])
|
|
|
|
>>> blocks3 = sc.parallelize([((0, 0), sm), ((1, 0), dm2)])
|
|
|
|
>>> mat1 = BlockMatrix(blocks1, 3, 2)
|
|
|
|
>>> mat2 = BlockMatrix(blocks2, 3, 2)
|
|
|
|
>>> mat3 = BlockMatrix(blocks3, 3, 2)
|
|
|
|
|
|
|
|
>>> mat1.subtract(mat2).toLocalMatrix()
|
|
|
|
DenseMatrix(6, 2, [-4.0, -7.0, -4.0, 4.0, 7.0, 4.0, -6.0, -5.0, -10.0, 6.0, 5.0, 10.0], 0)
|
|
|
|
|
|
|
|
>>> mat2.subtract(mat3).toLocalMatrix()
|
|
|
|
DenseMatrix(6, 2, [6.0, 8.0, 9.0, -4.0, -7.0, -4.0, 10.0, 9.0, 9.0, -6.0, -5.0, -10.0], 0)
|
|
|
|
"""
|
|
|
|
if not isinstance(other, BlockMatrix):
|
|
|
|
raise TypeError("Other should be a BlockMatrix, got %s" % type(other))
|
|
|
|
|
|
|
|
other_java_block_matrix = other._java_matrix_wrapper._java_model
|
|
|
|
java_block_matrix = self._java_matrix_wrapper.call("subtract", other_java_block_matrix)
|
|
|
|
return BlockMatrix(java_block_matrix, self.rowsPerBlock, self.colsPerBlock)
|
|
|
|
|
2015-10-27 14:05:14 -04:00
|
|
|
def multiply(self, other):
|
|
|
|
"""
|
|
|
|
Left multiplies this BlockMatrix by `other`, another
|
|
|
|
BlockMatrix. The `colsPerBlock` of this matrix must equal the
|
|
|
|
`rowsPerBlock` of `other`. If `other` contains any SparseMatrix
|
|
|
|
blocks, they will have to be converted to DenseMatrix blocks.
|
|
|
|
The output BlockMatrix will only consist of DenseMatrix blocks.
|
|
|
|
This may cause some performance issues until support for
|
|
|
|
multiplying two sparse matrices is added.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-10-27 14:05:14 -04:00
|
|
|
>>> dm1 = Matrices.dense(2, 3, [1, 2, 3, 4, 5, 6])
|
|
|
|
>>> dm2 = Matrices.dense(2, 3, [7, 8, 9, 10, 11, 12])
|
|
|
|
>>> dm3 = Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])
|
|
|
|
>>> dm4 = Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12])
|
|
|
|
>>> sm = Matrices.sparse(3, 2, [0, 1, 3], [0, 1, 2], [7, 11, 12])
|
|
|
|
>>> blocks1 = sc.parallelize([((0, 0), dm1), ((0, 1), dm2)])
|
|
|
|
>>> blocks2 = sc.parallelize([((0, 0), dm3), ((1, 0), dm4)])
|
|
|
|
>>> blocks3 = sc.parallelize([((0, 0), sm), ((1, 0), dm4)])
|
|
|
|
>>> mat1 = BlockMatrix(blocks1, 2, 3)
|
|
|
|
>>> mat2 = BlockMatrix(blocks2, 3, 2)
|
|
|
|
>>> mat3 = BlockMatrix(blocks3, 3, 2)
|
|
|
|
|
|
|
|
>>> mat1.multiply(mat2).toLocalMatrix()
|
|
|
|
DenseMatrix(2, 2, [242.0, 272.0, 350.0, 398.0], 0)
|
|
|
|
|
|
|
|
>>> mat1.multiply(mat3).toLocalMatrix()
|
|
|
|
DenseMatrix(2, 2, [227.0, 258.0, 394.0, 450.0], 0)
|
|
|
|
"""
|
|
|
|
if not isinstance(other, BlockMatrix):
|
|
|
|
raise TypeError("Other should be a BlockMatrix, got %s" % type(other))
|
|
|
|
|
|
|
|
other_java_block_matrix = other._java_matrix_wrapper._java_model
|
|
|
|
java_block_matrix = self._java_matrix_wrapper.call("multiply", other_java_block_matrix)
|
|
|
|
return BlockMatrix(java_block_matrix, self.rowsPerBlock, self.colsPerBlock)
|
|
|
|
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
def transpose(self):
|
|
|
|
"""
|
|
|
|
Transpose this BlockMatrix. Returns a new BlockMatrix
|
|
|
|
instance sharing the same underlying data. Is a lazy operation.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
.. versionadded:: 2.0.0
|
|
|
|
|
|
|
|
Examples
|
|
|
|
--------
|
[SPARK-9656][MLLIB][PYTHON] Add missing methods to PySpark's Distributed Linear Algebra Classes
This PR adds the remaining group of methods to PySpark's distributed linear algebra classes as follows:
* `RowMatrix` <sup>**[1]**</sup>
1. `computeGramianMatrix`
2. `computeCovariance`
3. `computeColumnSummaryStatistics`
4. `columnSimilarities`
5. `tallSkinnyQR` <sup>**[2]**</sup>
* `IndexedRowMatrix` <sup>**[3]**</sup>
1. `computeGramianMatrix`
* `CoordinateMatrix`
1. `transpose`
* `BlockMatrix`
1. `validate`
2. `cache`
3. `persist`
4. `transpose`
**[1]**: Note: `multiply`, `computeSVD`, and `computePrincipalComponents` are already part of PR #7963 for SPARK-6227.
**[2]**: Implementing `tallSkinnyQR` uncovered a bug with our PySpark `RowMatrix` constructor. As discussed on the dev list [here](http://apache-spark-developers-list.1001551.n3.nabble.com/K-Means-And-Class-Tags-td10038.html), there appears to be an issue with type erasure with RDDs coming from Java, and by extension from PySpark. Although we are attempting to construct a `RowMatrix` from an `RDD[Vector]` in [PythonMLlibAPI](https://github.com/apache/spark/blob/master/mllib/src/main/scala/org/apache/spark/mllib/api/python/PythonMLLibAPI.scala#L1115), the `Vector` type is erased, resulting in an `RDD[Object]`. Thus, when calling Scala's `tallSkinnyQR` from PySpark, we get a Java `ClassCastException` in which an `Object` cannot be cast to a Spark `Vector`. As noted in the aforementioned dev list thread, this issue was also encountered with `DecisionTrees`, and the fix involved an explicit `retag` of the RDD with a `Vector` type. Thus, this PR currently contains that fix applied to the `createRowMatrix` helper function in `PythonMLlibAPI`. `IndexedRowMatrix` and `CoordinateMatrix` do not appear to have this issue likely due to their related helper functions in `PythonMLlibAPI` creating the RDDs explicitly from DataFrames with pattern matching, thus preserving the types. However, this fix may be out of scope for this single PR, and it may be better suited in a separate JIRA/PR. Therefore, I have marked this PR as WIP and am open to discussion.
**[3]**: Note: `multiply` and `computeSVD` are already part of PR #7963 for SPARK-6227.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #9441 from dusenberrymw/SPARK-9656_Add_Missing_Methods_to_PySpark_Distributed_Linear_Algebra.
2016-04-27 13:48:05 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2)
|
|
|
|
|
|
|
|
>>> mat_transposed = mat.transpose()
|
|
|
|
>>> mat_transposed.toLocalMatrix()
|
|
|
|
DenseMatrix(2, 6, [1.0, 4.0, 2.0, 5.0, 3.0, 6.0, 7.0, 10.0, 8.0, 11.0, 9.0, 12.0], 0)
|
|
|
|
"""
|
|
|
|
java_transposed_matrix = self._java_matrix_wrapper.call("transpose")
|
|
|
|
return BlockMatrix(java_transposed_matrix, self.colsPerBlock, self.rowsPerBlock)
|
|
|
|
|
2015-08-05 10:40:50 -04:00
|
|
|
def toLocalMatrix(self):
|
|
|
|
"""
|
|
|
|
Collect the distributed matrix on the driver as a DenseMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2).toLocalMatrix()
|
|
|
|
|
|
|
|
>>> # This BlockMatrix will have 6 effective rows, due to
|
|
|
|
>>> # having two sub-matrix blocks stacked, each with 3 rows.
|
|
|
|
>>> # The ensuing DenseMatrix will also have 6 rows.
|
|
|
|
>>> print(mat.numRows)
|
|
|
|
6
|
|
|
|
|
|
|
|
>>> # This BlockMatrix will have 2 effective columns, due to
|
|
|
|
>>> # having two sub-matrix blocks stacked, each with 2
|
|
|
|
>>> # columns. The ensuing DenseMatrix will also have 2 columns.
|
|
|
|
>>> print(mat.numCols)
|
|
|
|
2
|
|
|
|
"""
|
|
|
|
return self._java_matrix_wrapper.call("toLocalMatrix")
|
|
|
|
|
|
|
|
def toIndexedRowMatrix(self):
|
|
|
|
"""
|
|
|
|
Convert this matrix to an IndexedRowMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(3, 2, [1, 2, 3, 4, 5, 6])),
|
|
|
|
... ((1, 0), Matrices.dense(3, 2, [7, 8, 9, 10, 11, 12]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 3, 2).toIndexedRowMatrix()
|
|
|
|
|
|
|
|
>>> # This BlockMatrix will have 6 effective rows, due to
|
|
|
|
>>> # having two sub-matrix blocks stacked, each with 3 rows.
|
|
|
|
>>> # The ensuing IndexedRowMatrix will also have 6 rows.
|
|
|
|
>>> print(mat.numRows())
|
|
|
|
6
|
|
|
|
|
|
|
|
>>> # This BlockMatrix will have 2 effective columns, due to
|
|
|
|
>>> # having two sub-matrix blocks stacked, each with 2 columns.
|
|
|
|
>>> # The ensuing IndexedRowMatrix will also have 2 columns.
|
|
|
|
>>> print(mat.numCols())
|
|
|
|
2
|
|
|
|
"""
|
|
|
|
java_indexed_row_matrix = self._java_matrix_wrapper.call("toIndexedRowMatrix")
|
|
|
|
return IndexedRowMatrix(java_indexed_row_matrix)
|
|
|
|
|
|
|
|
def toCoordinateMatrix(self):
|
|
|
|
"""
|
|
|
|
Convert this matrix to a CoordinateMatrix.
|
|
|
|
|
2020-11-24 20:24:41 -05:00
|
|
|
Examples
|
|
|
|
--------
|
2015-08-05 10:40:50 -04:00
|
|
|
>>> blocks = sc.parallelize([((0, 0), Matrices.dense(1, 2, [1, 2])),
|
|
|
|
... ((1, 0), Matrices.dense(1, 2, [7, 8]))])
|
|
|
|
>>> mat = BlockMatrix(blocks, 1, 2).toCoordinateMatrix()
|
|
|
|
>>> mat.entries.take(3)
|
|
|
|
[MatrixEntry(0, 0, 1.0), MatrixEntry(0, 1, 2.0), MatrixEntry(1, 0, 7.0)]
|
|
|
|
"""
|
|
|
|
java_coordinate_matrix = self._java_matrix_wrapper.call("toCoordinateMatrix")
|
|
|
|
return CoordinateMatrix(java_coordinate_matrix)
|
|
|
|
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
def _test():
|
|
|
|
import doctest
|
[SPARK-24740][PYTHON][ML] Make PySpark's tests compatible with NumPy 1.14+
## What changes were proposed in this pull request?
This PR proposes to make PySpark's tests compatible with NumPy 0.14+
NumPy 0.14.x introduced rather radical changes about its string representation.
For example, the tests below are failed:
```
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 895, in __main__.DenseMatrix.__str__
Failed example:
print(dm)
Expected:
DenseMatrix([[ 0., 2.],
[ 1., 3.]])
Got:
DenseMatrix([[0., 2.],
[1., 3.]])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 899, in __main__.DenseMatrix.__str__
Failed example:
print(dm)
Expected:
DenseMatrix([[ 0., 1.],
[ 2., 3.]])
Got:
DenseMatrix([[0., 1.],
[2., 3.]])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 939, in __main__.DenseMatrix.toArray
Failed example:
m.toArray()
Expected:
array([[ 0., 2.],
[ 1., 3.]])
Got:
array([[0., 2.],
[1., 3.]])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 324, in __main__.DenseVector.dot
Failed example:
dense.dot(np.reshape([1., 2., 3., 4.], (2, 2), order='F'))
Expected:
array([ 5., 11.])
Got:
array([ 5., 11.])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 567, in __main__.SparseVector.dot
Failed example:
a.dot(np.array([[1, 1], [2, 2], [3, 3], [4, 4]]))
Expected:
array([ 22., 22.])
Got:
array([22., 22.])
```
See [release note](https://docs.scipy.org/doc/numpy-1.14.0/release.html#compatibility-notes).
## How was this patch tested?
Manually tested:
```
$ ./run-tests --python-executables=python3.6,python2.7 --modules=pyspark-ml,pyspark-mllib
Running PySpark tests. Output is in /.../spark/python/unit-tests.log
Will test against the following Python executables: ['python3.6', 'python2.7']
Will test the following Python modules: ['pyspark-ml', 'pyspark-mllib']
Starting test(python2.7): pyspark.mllib.tests
Starting test(python2.7): pyspark.ml.classification
Starting test(python3.6): pyspark.mllib.tests
Starting test(python2.7): pyspark.ml.clustering
Finished test(python2.7): pyspark.ml.clustering (54s)
Starting test(python2.7): pyspark.ml.evaluation
Finished test(python2.7): pyspark.ml.classification (74s)
Starting test(python2.7): pyspark.ml.feature
Finished test(python2.7): pyspark.ml.evaluation (27s)
Starting test(python2.7): pyspark.ml.fpm
Finished test(python2.7): pyspark.ml.fpm (0s)
Starting test(python2.7): pyspark.ml.image
Finished test(python2.7): pyspark.ml.image (17s)
Starting test(python2.7): pyspark.ml.linalg.__init__
Finished test(python2.7): pyspark.ml.linalg.__init__ (1s)
Starting test(python2.7): pyspark.ml.recommendation
Finished test(python2.7): pyspark.ml.feature (76s)
Starting test(python2.7): pyspark.ml.regression
Finished test(python2.7): pyspark.ml.recommendation (69s)
Starting test(python2.7): pyspark.ml.stat
Finished test(python2.7): pyspark.ml.regression (45s)
Starting test(python2.7): pyspark.ml.tests
Finished test(python2.7): pyspark.ml.stat (28s)
Starting test(python2.7): pyspark.ml.tuning
Finished test(python2.7): pyspark.ml.tuning (20s)
Starting test(python2.7): pyspark.mllib.classification
Finished test(python2.7): pyspark.mllib.classification (31s)
Starting test(python2.7): pyspark.mllib.clustering
Finished test(python2.7): pyspark.mllib.tests (260s)
Starting test(python2.7): pyspark.mllib.evaluation
Finished test(python3.6): pyspark.mllib.tests (266s)
Starting test(python2.7): pyspark.mllib.feature
Finished test(python2.7): pyspark.mllib.evaluation (21s)
Starting test(python2.7): pyspark.mllib.fpm
Finished test(python2.7): pyspark.mllib.feature (38s)
Starting test(python2.7): pyspark.mllib.linalg.__init__
Finished test(python2.7): pyspark.mllib.linalg.__init__ (1s)
Starting test(python2.7): pyspark.mllib.linalg.distributed
Finished test(python2.7): pyspark.mllib.fpm (34s)
Starting test(python2.7): pyspark.mllib.random
Finished test(python2.7): pyspark.mllib.clustering (64s)
Starting test(python2.7): pyspark.mllib.recommendation
Finished test(python2.7): pyspark.mllib.random (15s)
Starting test(python2.7): pyspark.mllib.regression
Finished test(python2.7): pyspark.mllib.linalg.distributed (47s)
Starting test(python2.7): pyspark.mllib.stat.KernelDensity
Finished test(python2.7): pyspark.mllib.stat.KernelDensity (0s)
Starting test(python2.7): pyspark.mllib.stat._statistics
Finished test(python2.7): pyspark.mllib.recommendation (40s)
Starting test(python2.7): pyspark.mllib.tree
Finished test(python2.7): pyspark.mllib.regression (38s)
Starting test(python2.7): pyspark.mllib.util
Finished test(python2.7): pyspark.mllib.stat._statistics (19s)
Starting test(python3.6): pyspark.ml.classification
Finished test(python2.7): pyspark.mllib.tree (26s)
Starting test(python3.6): pyspark.ml.clustering
Finished test(python2.7): pyspark.mllib.util (27s)
Starting test(python3.6): pyspark.ml.evaluation
Finished test(python3.6): pyspark.ml.evaluation (30s)
Starting test(python3.6): pyspark.ml.feature
Finished test(python2.7): pyspark.ml.tests (234s)
Starting test(python3.6): pyspark.ml.fpm
Finished test(python3.6): pyspark.ml.fpm (1s)
Starting test(python3.6): pyspark.ml.image
Finished test(python3.6): pyspark.ml.clustering (55s)
Starting test(python3.6): pyspark.ml.linalg.__init__
Finished test(python3.6): pyspark.ml.linalg.__init__ (0s)
Starting test(python3.6): pyspark.ml.recommendation
Finished test(python3.6): pyspark.ml.classification (71s)
Starting test(python3.6): pyspark.ml.regression
Finished test(python3.6): pyspark.ml.image (18s)
Starting test(python3.6): pyspark.ml.stat
Finished test(python3.6): pyspark.ml.stat (37s)
Starting test(python3.6): pyspark.ml.tests
Finished test(python3.6): pyspark.ml.regression (59s)
Starting test(python3.6): pyspark.ml.tuning
Finished test(python3.6): pyspark.ml.feature (93s)
Starting test(python3.6): pyspark.mllib.classification
Finished test(python3.6): pyspark.ml.recommendation (83s)
Starting test(python3.6): pyspark.mllib.clustering
Finished test(python3.6): pyspark.ml.tuning (29s)
Starting test(python3.6): pyspark.mllib.evaluation
Finished test(python3.6): pyspark.mllib.evaluation (26s)
Starting test(python3.6): pyspark.mllib.feature
Finished test(python3.6): pyspark.mllib.classification (43s)
Starting test(python3.6): pyspark.mllib.fpm
Finished test(python3.6): pyspark.mllib.clustering (81s)
Starting test(python3.6): pyspark.mllib.linalg.__init__
Finished test(python3.6): pyspark.mllib.linalg.__init__ (2s)
Starting test(python3.6): pyspark.mllib.linalg.distributed
Finished test(python3.6): pyspark.mllib.fpm (48s)
Starting test(python3.6): pyspark.mllib.random
Finished test(python3.6): pyspark.mllib.feature (54s)
Starting test(python3.6): pyspark.mllib.recommendation
Finished test(python3.6): pyspark.mllib.random (18s)
Starting test(python3.6): pyspark.mllib.regression
Finished test(python3.6): pyspark.mllib.linalg.distributed (55s)
Starting test(python3.6): pyspark.mllib.stat.KernelDensity
Finished test(python3.6): pyspark.mllib.stat.KernelDensity (1s)
Starting test(python3.6): pyspark.mllib.stat._statistics
Finished test(python3.6): pyspark.mllib.recommendation (51s)
Starting test(python3.6): pyspark.mllib.tree
Finished test(python3.6): pyspark.mllib.regression (45s)
Starting test(python3.6): pyspark.mllib.util
Finished test(python3.6): pyspark.mllib.stat._statistics (21s)
Finished test(python3.6): pyspark.mllib.tree (27s)
Finished test(python3.6): pyspark.mllib.util (27s)
Finished test(python3.6): pyspark.ml.tests (264s)
```
Author: hyukjinkwon <gurwls223@apache.org>
Closes #21715 from HyukjinKwon/SPARK-24740.
2018-07-06 23:39:29 -04:00
|
|
|
import numpy
|
2016-05-23 21:14:48 -04:00
|
|
|
from pyspark.sql import SparkSession
|
2015-08-05 10:40:50 -04:00
|
|
|
from pyspark.mllib.linalg import Matrices
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
import pyspark.mllib.linalg.distributed
|
[SPARK-24740][PYTHON][ML] Make PySpark's tests compatible with NumPy 1.14+
## What changes were proposed in this pull request?
This PR proposes to make PySpark's tests compatible with NumPy 0.14+
NumPy 0.14.x introduced rather radical changes about its string representation.
For example, the tests below are failed:
```
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 895, in __main__.DenseMatrix.__str__
Failed example:
print(dm)
Expected:
DenseMatrix([[ 0., 2.],
[ 1., 3.]])
Got:
DenseMatrix([[0., 2.],
[1., 3.]])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 899, in __main__.DenseMatrix.__str__
Failed example:
print(dm)
Expected:
DenseMatrix([[ 0., 1.],
[ 2., 3.]])
Got:
DenseMatrix([[0., 1.],
[2., 3.]])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 939, in __main__.DenseMatrix.toArray
Failed example:
m.toArray()
Expected:
array([[ 0., 2.],
[ 1., 3.]])
Got:
array([[0., 2.],
[1., 3.]])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 324, in __main__.DenseVector.dot
Failed example:
dense.dot(np.reshape([1., 2., 3., 4.], (2, 2), order='F'))
Expected:
array([ 5., 11.])
Got:
array([ 5., 11.])
**********************************************************************
File "/.../spark/python/pyspark/ml/linalg/__init__.py", line 567, in __main__.SparseVector.dot
Failed example:
a.dot(np.array([[1, 1], [2, 2], [3, 3], [4, 4]]))
Expected:
array([ 22., 22.])
Got:
array([22., 22.])
```
See [release note](https://docs.scipy.org/doc/numpy-1.14.0/release.html#compatibility-notes).
## How was this patch tested?
Manually tested:
```
$ ./run-tests --python-executables=python3.6,python2.7 --modules=pyspark-ml,pyspark-mllib
Running PySpark tests. Output is in /.../spark/python/unit-tests.log
Will test against the following Python executables: ['python3.6', 'python2.7']
Will test the following Python modules: ['pyspark-ml', 'pyspark-mllib']
Starting test(python2.7): pyspark.mllib.tests
Starting test(python2.7): pyspark.ml.classification
Starting test(python3.6): pyspark.mllib.tests
Starting test(python2.7): pyspark.ml.clustering
Finished test(python2.7): pyspark.ml.clustering (54s)
Starting test(python2.7): pyspark.ml.evaluation
Finished test(python2.7): pyspark.ml.classification (74s)
Starting test(python2.7): pyspark.ml.feature
Finished test(python2.7): pyspark.ml.evaluation (27s)
Starting test(python2.7): pyspark.ml.fpm
Finished test(python2.7): pyspark.ml.fpm (0s)
Starting test(python2.7): pyspark.ml.image
Finished test(python2.7): pyspark.ml.image (17s)
Starting test(python2.7): pyspark.ml.linalg.__init__
Finished test(python2.7): pyspark.ml.linalg.__init__ (1s)
Starting test(python2.7): pyspark.ml.recommendation
Finished test(python2.7): pyspark.ml.feature (76s)
Starting test(python2.7): pyspark.ml.regression
Finished test(python2.7): pyspark.ml.recommendation (69s)
Starting test(python2.7): pyspark.ml.stat
Finished test(python2.7): pyspark.ml.regression (45s)
Starting test(python2.7): pyspark.ml.tests
Finished test(python2.7): pyspark.ml.stat (28s)
Starting test(python2.7): pyspark.ml.tuning
Finished test(python2.7): pyspark.ml.tuning (20s)
Starting test(python2.7): pyspark.mllib.classification
Finished test(python2.7): pyspark.mllib.classification (31s)
Starting test(python2.7): pyspark.mllib.clustering
Finished test(python2.7): pyspark.mllib.tests (260s)
Starting test(python2.7): pyspark.mllib.evaluation
Finished test(python3.6): pyspark.mllib.tests (266s)
Starting test(python2.7): pyspark.mllib.feature
Finished test(python2.7): pyspark.mllib.evaluation (21s)
Starting test(python2.7): pyspark.mllib.fpm
Finished test(python2.7): pyspark.mllib.feature (38s)
Starting test(python2.7): pyspark.mllib.linalg.__init__
Finished test(python2.7): pyspark.mllib.linalg.__init__ (1s)
Starting test(python2.7): pyspark.mllib.linalg.distributed
Finished test(python2.7): pyspark.mllib.fpm (34s)
Starting test(python2.7): pyspark.mllib.random
Finished test(python2.7): pyspark.mllib.clustering (64s)
Starting test(python2.7): pyspark.mllib.recommendation
Finished test(python2.7): pyspark.mllib.random (15s)
Starting test(python2.7): pyspark.mllib.regression
Finished test(python2.7): pyspark.mllib.linalg.distributed (47s)
Starting test(python2.7): pyspark.mllib.stat.KernelDensity
Finished test(python2.7): pyspark.mllib.stat.KernelDensity (0s)
Starting test(python2.7): pyspark.mllib.stat._statistics
Finished test(python2.7): pyspark.mllib.recommendation (40s)
Starting test(python2.7): pyspark.mllib.tree
Finished test(python2.7): pyspark.mllib.regression (38s)
Starting test(python2.7): pyspark.mllib.util
Finished test(python2.7): pyspark.mllib.stat._statistics (19s)
Starting test(python3.6): pyspark.ml.classification
Finished test(python2.7): pyspark.mllib.tree (26s)
Starting test(python3.6): pyspark.ml.clustering
Finished test(python2.7): pyspark.mllib.util (27s)
Starting test(python3.6): pyspark.ml.evaluation
Finished test(python3.6): pyspark.ml.evaluation (30s)
Starting test(python3.6): pyspark.ml.feature
Finished test(python2.7): pyspark.ml.tests (234s)
Starting test(python3.6): pyspark.ml.fpm
Finished test(python3.6): pyspark.ml.fpm (1s)
Starting test(python3.6): pyspark.ml.image
Finished test(python3.6): pyspark.ml.clustering (55s)
Starting test(python3.6): pyspark.ml.linalg.__init__
Finished test(python3.6): pyspark.ml.linalg.__init__ (0s)
Starting test(python3.6): pyspark.ml.recommendation
Finished test(python3.6): pyspark.ml.classification (71s)
Starting test(python3.6): pyspark.ml.regression
Finished test(python3.6): pyspark.ml.image (18s)
Starting test(python3.6): pyspark.ml.stat
Finished test(python3.6): pyspark.ml.stat (37s)
Starting test(python3.6): pyspark.ml.tests
Finished test(python3.6): pyspark.ml.regression (59s)
Starting test(python3.6): pyspark.ml.tuning
Finished test(python3.6): pyspark.ml.feature (93s)
Starting test(python3.6): pyspark.mllib.classification
Finished test(python3.6): pyspark.ml.recommendation (83s)
Starting test(python3.6): pyspark.mllib.clustering
Finished test(python3.6): pyspark.ml.tuning (29s)
Starting test(python3.6): pyspark.mllib.evaluation
Finished test(python3.6): pyspark.mllib.evaluation (26s)
Starting test(python3.6): pyspark.mllib.feature
Finished test(python3.6): pyspark.mllib.classification (43s)
Starting test(python3.6): pyspark.mllib.fpm
Finished test(python3.6): pyspark.mllib.clustering (81s)
Starting test(python3.6): pyspark.mllib.linalg.__init__
Finished test(python3.6): pyspark.mllib.linalg.__init__ (2s)
Starting test(python3.6): pyspark.mllib.linalg.distributed
Finished test(python3.6): pyspark.mllib.fpm (48s)
Starting test(python3.6): pyspark.mllib.random
Finished test(python3.6): pyspark.mllib.feature (54s)
Starting test(python3.6): pyspark.mllib.recommendation
Finished test(python3.6): pyspark.mllib.random (18s)
Starting test(python3.6): pyspark.mllib.regression
Finished test(python3.6): pyspark.mllib.linalg.distributed (55s)
Starting test(python3.6): pyspark.mllib.stat.KernelDensity
Finished test(python3.6): pyspark.mllib.stat.KernelDensity (1s)
Starting test(python3.6): pyspark.mllib.stat._statistics
Finished test(python3.6): pyspark.mllib.recommendation (51s)
Starting test(python3.6): pyspark.mllib.tree
Finished test(python3.6): pyspark.mllib.regression (45s)
Starting test(python3.6): pyspark.mllib.util
Finished test(python3.6): pyspark.mllib.stat._statistics (21s)
Finished test(python3.6): pyspark.mllib.tree (27s)
Finished test(python3.6): pyspark.mllib.util (27s)
Finished test(python3.6): pyspark.ml.tests (264s)
```
Author: hyukjinkwon <gurwls223@apache.org>
Closes #21715 from HyukjinKwon/SPARK-24740.
2018-07-06 23:39:29 -04:00
|
|
|
try:
|
|
|
|
# Numpy 1.14+ changed it's string format.
|
|
|
|
numpy.set_printoptions(legacy='1.13')
|
|
|
|
except TypeError:
|
|
|
|
pass
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
globs = pyspark.mllib.linalg.distributed.__dict__.copy()
|
2016-05-23 21:14:48 -04:00
|
|
|
spark = SparkSession.builder\
|
|
|
|
.master("local[2]")\
|
|
|
|
.appName("mllib.linalg.distributed tests")\
|
|
|
|
.getOrCreate()
|
|
|
|
globs['sc'] = spark.sparkContext
|
2015-08-05 10:40:50 -04:00
|
|
|
globs['Matrices'] = Matrices
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
(failure_count, test_count) = doctest.testmod(globs=globs, optionflags=doctest.ELLIPSIS)
|
2016-05-23 21:14:48 -04:00
|
|
|
spark.stop()
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
if failure_count:
|
2018-03-08 06:38:34 -05:00
|
|
|
sys.exit(-1)
|
[SPARK-6485] [MLLIB] [PYTHON] Add CoordinateMatrix/RowMatrix/IndexedRowMatrix to PySpark.
This PR adds the RowMatrix, IndexedRowMatrix, and CoordinateMatrix distributed matrices to PySpark. Each distributed matrix class acts as a wrapper around the Scala/Java counterpart by maintaining a reference to the Java object. New distributed matrices can be created using factory methods added to DistributedMatrices, which creates the Java distributed matrix and then wraps it with the corresponding PySpark class. This design allows for simple conversion between the various distributed matrices, and lets us re-use the Scala code. Serialization between Python and Java is implemented using DataFrames as needed for IndexedRowMatrix and CoordinateMatrix for simplicity. Associated documentation and unit-tests have also been added. To facilitate code review, this PR implements access to the rows/entries as RDDs, the number of rows & columns, and conversions between the various distributed matrices (not including BlockMatrix), and does not implement the other linear algebra functions of the matrices, although this will be very simple to add now.
Author: Mike Dusenberry <mwdusenb@us.ibm.com>
Closes #7554 from dusenberrymw/SPARK-6485_Add_CoordinateMatrix_RowMatrix_IndexedMatrix_to_PySpark and squashes the following commits:
bb039cb [Mike Dusenberry] Minor documentation update.
b887c18 [Mike Dusenberry] Updating the matrix conversion logic again to make it even cleaner. Now, we allow the 'rows' parameter in the constructors to be either an RDD or the Java matrix object. If 'rows' is an RDD, we create a Java matrix object, wrap it, and then store that. If 'rows' is a Java matrix object of the correct type, we just wrap and store that directly. This is only for internal usage, and publicly, we still require 'rows' to be an RDD. We no longer store the 'rows' RDD, and instead just compute it from the Java object when needed. The point of this is that when we do matrix conversions, we do the conversion on the Scala/Java side, which returns a Java object, so we should use that directly, but exposing 'java_matrix' parameter in the public API is not ideal. This non-public feature of allowing 'rows' to be a Java matrix object is documented in the '__init__' constructor docstrings, which are not part of the generated public API, and doctests are also included.
7f0dcb6 [Mike Dusenberry] Updating module docstring.
cfc1be5 [Mike Dusenberry] Use 'new SQLContext(matrix.rows.sparkContext)' rather than 'SQLContext.getOrCreate', as the later doesn't guarantee that the SparkContext will be the same as for the matrix.rows data.
687e345 [Mike Dusenberry] Improving conversion performance. This adds an optional 'java_matrix' parameter to the constructors, and pulls the conversion logic out into a '_create_from_java' function. Now, if the constructors are given a valid Java distributed matrix object as 'java_matrix', they will store those internally, rather than create a new one on the Scala/Java side.
3e50b6e [Mike Dusenberry] Moving the distributed matrices to pyspark.mllib.linalg.distributed.
308f197 [Mike Dusenberry] Using properties for better documentation.
1633f86 [Mike Dusenberry] Minor documentation cleanup.
f0c13a7 [Mike Dusenberry] CoordinateMatrix should inherit from DistributedMatrix.
ffdd724 [Mike Dusenberry] Updating doctests to make documentation cleaner.
3fd4016 [Mike Dusenberry] Updating docstrings.
27cd5f6 [Mike Dusenberry] Simplifying input conversions in the constructors for each distributed matrix.
a409cf5 [Mike Dusenberry] Updating doctests to be less verbose by using lists instead of DenseVectors explicitly.
d19b0ba [Mike Dusenberry] Updating code and documentation to note that a vector-like object (numpy array, list, etc.) can be used in place of explicit Vector object, and adding conversions when necessary to RowMatrix construction.
4bd756d [Mike Dusenberry] Adding param documentation to IndexedRow and MatrixEntry.
c6bded5 [Mike Dusenberry] Move conversion logic from tuples to IndexedRow or MatrixEntry types from within the IndexedRowMatrix and CoordinateMatrix constructors to separate _convert_to_indexed_row and _convert_to_matrix_entry functions.
329638b [Mike Dusenberry] Moving the Experimental tag to the top of each docstring.
0be6826 [Mike Dusenberry] Simplifying doctests by removing duplicated rows/entries RDDs within the various tests.
c0900df [Mike Dusenberry] Adding the colons that were accidentally not inserted.
4ad6819 [Mike Dusenberry] Documenting the and parameters.
3b854b9 [Mike Dusenberry] Minor updates to documentation.
10046e8 [Mike Dusenberry] Updating documentation to use class constructors instead of the removed DistributedMatrices factory methods.
119018d [Mike Dusenberry] Adding static methods to each of the distributed matrix classes to consolidate conversion logic.
4d7af86 [Mike Dusenberry] Adding type checks to the constructors. Although it is slightly verbose, it is better for the user to have a good error message than a cryptic stacktrace.
93b6a3d [Mike Dusenberry] Pulling the DistributedMatrices Python class out of this pull request.
f6f3c68 [Mike Dusenberry] Pulling the DistributedMatrices Scala class out of this pull request.
6a3ecb7 [Mike Dusenberry] Updating pattern matching.
08f287b [Mike Dusenberry] Slight reformatting of the documentation.
a245dc0 [Mike Dusenberry] Updating Python doctests for compatability between Python 2 & 3. Since Python 3 removed the idea of a separate 'long' type, all values that would have been outputted as a 'long' (ex: '4L') will now be treated as an 'int' and outputed as one (ex: '4'). The doctests now explicitly convert to ints so that both Python 2 and 3 will have the same output. This is fine since the values are all small, and thus can be easily represented as ints.
4d3a37e [Mike Dusenberry] Reformatting a few long Python doctest lines.
7e3ca16 [Mike Dusenberry] Fixing long lines.
f721ead [Mike Dusenberry] Updating documentation for each of the distributed matrices.
ab0e8b6 [Mike Dusenberry] Updating unit test to be more useful.
dda2f89 [Mike Dusenberry] Added wrappers for the conversions between the various distributed matrices. Added logic to be able to access the rows/entries of the distributed matrices, which requires serialization through DataFrames for IndexedRowMatrix and CoordinateMatrix types. Added unit tests.
0cd7166 [Mike Dusenberry] Implemented the CoordinateMatrix API in PySpark, following the idea of the IndexedRowMatrix API, including using DataFrames for serialization.
3c369cb [Mike Dusenberry] Updating the architecture a bit to make conversions between the various distributed matrix types easier. The different distributed matrix classes are now only wrappers around the Java objects, and take the Java object as an argument during construction. This way, we can call for example on an , which returns a reference to a Java RowMatrix object, and then construct a PySpark RowMatrix object wrapped around the Java object. This is analogous to the behavior of PySpark RDDs and DataFrames. We now delegate creation of the various distributed matrices from scratch in PySpark to the factory methods on .
4bdd09b [Mike Dusenberry] Implemented the IndexedRowMatrix API in PySpark, following the idea of the RowMatrix API. Note that for the IndexedRowMatrix, we use DataFrames to serialize the data between Python and Scala/Java, so we accept PySpark RDDs, then convert to a DataFrame, then convert back to RDDs on the Scala/Java side before constructing the IndexedRowMatrix.
23bf1ec [Mike Dusenberry] Updating documentation to add PySpark RowMatrix. Inserting newline above doctest so that it renders properly in API docs.
b194623 [Mike Dusenberry] Updating design to have a PySpark RowMatrix simply create and keep a reference to a wrapper over a Java RowMatrix. Updating DistributedMatrices factory methods to accept numRows and numCols with default values. Updating PySpark DistributedMatrices factory method to simply create a PySpark RowMatrix. Adding additional doctests for numRows and numCols parameters.
bc2d220 [Mike Dusenberry] Adding unit tests for RowMatrix methods.
d7e316f [Mike Dusenberry] Implemented the RowMatrix API in PySpark by doing the following: Added a DistributedMatrices class to contain factory methods for creating the various distributed matrices. Added a factory method for creating a RowMatrix from an RDD of Vectors. Added a createRowMatrix function to the PythonMLlibAPI to interface with the factory method. Added DistributedMatrix, DistributedMatrices, and RowMatrix classes to the pyspark.mllib.linalg api.
2015-08-04 19:30:03 -04:00
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
_test()
|