Floating point math is not exact, and most floating-point numbers end up being slightly imprecise due to rounding errors.
Simple values like 0.1 cannot be precisely represented using binary floating point numbers, and the limited precision of floating point numbers means that slight changes in the order of operations or the precision of intermediates can change the result.
That means that comparing two floats to see if they are equal is usually not what we want. As long as this imprecision stays small, it can usually be ignored.
Based on discussion in the community, we have implemented two different APIs for relative tolerance, and absolute tolerance. It makes sense that test writers should know which one they need depending on their circumstances.
Developers also need to explicitly specify the eps, and there is no default value which will sometimes cause confusion.
When comparing against zero using relative tolerance, a exception will be raised to warn users that it's meaningless.
For relative tolerance, users can now write
assert(23.1 ~== 23.52 relTol 0.02)
assert(23.1 ~== 22.74 relTol 0.02)
assert(23.1 ~= 23.52 relTol 0.02)
assert(23.1 ~= 22.74 relTol 0.02)
assert(!(23.1 !~= 23.52 relTol 0.02))
assert(!(23.1 !~= 22.74 relTol 0.02))
// This will throw exception with the following message.
// "Did not expect 23.1 and 23.52 to be within 0.02 using relative tolerance."
assert(23.1 !~== 23.52 relTol 0.02)
// "Expected 23.1 and 22.34 to be within 0.02 using relative tolerance."
assert(23.1 ~== 22.34 relTol 0.02)
For absolute error,
assert(17.8 ~== 17.99 absTol 0.2)
assert(17.8 ~== 17.61 absTol 0.2)
assert(17.8 ~= 17.99 absTol 0.2)
assert(17.8 ~= 17.61 absTol 0.2)
assert(!(17.8 !~= 17.99 absTol 0.2))
assert(!(17.8 !~= 17.61 absTol 0.2))
// This will throw exception with the following message.
// "Did not expect 17.8 and 17.99 to be within 0.2 using absolute error."
assert(17.8 !~== 17.99 absTol 0.2)
// "Expected 17.8 and 17.59 to be within 0.2 using absolute error."
assert(17.8 ~== 17.59 absTol 0.2)
Authors:
DB Tsai <dbtsaialpinenow.com>
Marek Kolodziej <marekalpinenow.com>
Author: DB Tsai <dbtsai@alpinenow.com>
Closes#1425 from dbtsai/SPARK-2479_comparing_floating_point and squashes the following commits:
8c7cbcc [DB Tsai] Alpine Data Labs
Utilities for generating random RDDs.
RandomRDD and RandomVectorRDD are created instead of using `sc.parallelize(range:Range)` because `Range` objects in Scala can only have `size <= Int.MaxValue`.
The object `RandomRDDGenerators` can be transformed into a generator class to reduce the number of auxiliary methods for optional arguments.
Author: Doris Xin <doris.s.xin@gmail.com>
Closes#1520 from dorx/randomRDD and squashes the following commits:
01121ac [Doris Xin] reviewer comments
6bf27d8 [Doris Xin] Merge branch 'master' into randomRDD
a8ea92d [Doris Xin] Reviewer comments
063ea0b [Doris Xin] Merge branch 'master' into randomRDD
aec68eb [Doris Xin] newline
bc90234 [Doris Xin] units passed.
d56cacb [Doris Xin] impl with RandomRDD
92d6f1c [Doris Xin] solution for Cloneable
df5bcff [Doris Xin] Merge branch 'generator' into randomRDD
f46d928 [Doris Xin] WIP
49ed20d [Doris Xin] alternative poisson distribution generator
7cb0e40 [Doris Xin] fix for data inconsistency
8881444 [Doris Xin] RandomRDDGenerator: initial design
Added a set of serializer/deserializer for Double in _common.py and PythonMLLibAPI in MLLib.
Author: Doris Xin <doris.s.xin@gmail.com>
Closes#1581 from dorx/doubleSerDe and squashes the following commits:
86a85b3 [Doris Xin] Merge branch 'master' into doubleSerDe
2bfe7a4 [Doris Xin] Removed magic byte
ad4d0d9 [Doris Xin] removed a space in unit
a9020bc [Doris Xin] units passed
7dad9af [Doris Xin] WIP
We saw task serialization problems with large feature dimension, which could be avoid if we don't serialize data directly into task but use broadcast variables. This PR uses broadcast in both training and prediction and adds tests to make sure the task size is small.
Author: Xiangrui Meng <meng@databricks.com>
Closes#1427 from mengxr/broadcast-new and squashes the following commits:
b9a1228 [Xiangrui Meng] style update
b97c184 [Xiangrui Meng] minimal change to LBFGS
9ebadcc [Xiangrui Meng] add task size test to RowMatrix
9427bf0 [Xiangrui Meng] add task size tests to linear methods
e0a5cf2 [Xiangrui Meng] add task size test to GD
28a8411 [Xiangrui Meng] add test for NaiveBayes
380778c [Xiangrui Meng] update KMeans test
bccab92 [Xiangrui Meng] add task size test to LBFGS
02103ba [Xiangrui Meng] remove print
e73d68e [Xiangrui Meng] update tests for k-means
174cb15 [Xiangrui Meng] use local-cluster for test with a small akka.frameSize
1928a5a [Xiangrui Meng] add test for KMeans task size
e00c2da [Xiangrui Meng] use broadcast in GD, KMeans
010d076 [Xiangrui Meng] modify NaiveBayesModel and GLM to use broadcast
JIRA: https://issues.apache.org/jira/browse/SPARK-2657
Our current code uses ArrayBuffers for each group of values in groupBy, as well as for the key's elements in CoGroupedRDD. ArrayBuffers have a lot of overhead if there are few values in them, which is likely to happen in cases such as join. In particular, they have a pointer to an Object[] of size 16 by default, which is 24 bytes for the array header + 128 for the pointers in there, plus at least 32 for the ArrayBuffer data structure. This patch replaces the per-group buffers with a CompactBuffer class that can store up to 2 elements more efficiently (in fields of itself) and acts like an ArrayBuffer beyond that. For a key's elements in CoGroupedRDD, we use an Array of CompactBuffers instead of an ArrayBuffer of ArrayBuffers.
There are some changes throughout the code to deal with CoGroupedRDD returning Array instead. We can also decide not to do that but CoGroupedRDD is a `DeveloperAPI` so I think it's okay to change it here.
Author: Matei Zaharia <matei@databricks.com>
Closes#1555 from mateiz/compact-groupby and squashes the following commits:
845a356 [Matei Zaharia] Lower initial size of CompactBuffer's vector to 8
07621a7 [Matei Zaharia] Review comments
0c1cd12 [Matei Zaharia] Don't use varargs in CompactBuffer.apply
bdc8a39 [Matei Zaharia] Small tweak to +=, and typos
f61f040 [Matei Zaharia] Fix line lengths
59da88b0 [Matei Zaharia] Fix line lengths
197cde8 [Matei Zaharia] Make CompactBuffer extend Seq to make its toSeq more efficient
775110f [Matei Zaharia] Change CoGroupedRDD to give (K, Array[Iterable[_]]) to avoid wrappers
9b4c6e8 [Matei Zaharia] Use CompactBuffer in CoGroupedRDD
ed577ab [Matei Zaharia] Use CompactBuffer in groupByKey
10f0de1 [Matei Zaharia] A CompactBuffer that's more memory-efficient than ArrayBuffer for small buffers
Allow small errors in comparison.
@dbtsai , this unit test blocks https://github.com/apache/spark/pull/1562 . I may need to merge this one first. We can change it to use the tools in https://github.com/apache/spark/pull/1425 after that PR gets merged.
Author: Xiangrui Meng <meng@databricks.com>
Closes#1576 from mengxr/fix-binary-metrics-unit-tests and squashes the following commits:
5076a7f [Xiangrui Meng] fix binary metrics unit tests
The name `preservesPartitioning` is ambiguous: 1) preserves the indices of partitions, 2) preserves the partitioner. The latter is correct and `preservesPartitioning` should really be called `preservesPartitioner` to avoid confusion. Unfortunately, this is already part of the API and we cannot change. We should be clear in the doc and fix wrong usages.
This PR
1. adds notes in `maPartitions*`,
2. makes `RDD.sample` preserve partitioner,
3. changes `preservesPartitioning` to false in `RDD.zip` because the keys of the first RDD are no longer the keys of the zipped RDD,
4. fixes some wrong usages in MLlib.
Author: Xiangrui Meng <meng@databricks.com>
Closes#1526 from mengxr/preserve-partitioner and squashes the following commits:
b361e65 [Xiangrui Meng] update doc based on pwendell's comments
3b1ba19 [Xiangrui Meng] update doc
357575c [Xiangrui Meng] fix unit test
20b4816 [Xiangrui Meng] Merge branch 'master' into preserve-partitioner
d1caa65 [Xiangrui Meng] add doc to explain preservesPartitioning fix wrong usage of preservesPartitioning make sample preserse partitioning
Author: peng.zhang <peng.zhang@xiaomi.com>
Closes#1521 from renozhang/fix-als and squashes the following commits:
b5727a4 [peng.zhang] Remove no need argument
1a4f7a0 [peng.zhang] Fix data skew in ALS
This is part of SPARK-2495 to allow users construct linear models manually.
Author: Xiangrui Meng <meng@databricks.com>
Closes#1492 from mengxr/public-constructor and squashes the following commits:
a48b766 [Xiangrui Meng] remove private[mllib] from linear models' constructors
The ability to perform multiclass classification is a big advantage for using decision trees and was a highly requested feature for mllib. This pull request adds multiclass classification support to the MLlib decision tree. It also adds sample weights support using WeightedLabeledPoint class for handling unbalanced datasets during classification. It will also support algorithms such as AdaBoost which requires instances to be weighted.
It handles the special case where the categorical variables cannot be ordered for multiclass classification and thus the optimizations used for speeding up binary classification cannot be directly used for multiclass classification with categorical variables. More specifically, for m categories in a categorical feature, it analyses all the ```2^(m-1) - 1``` categorical splits provided that #splits are less than the maxBins provided in the input. This condition will not be met for features with large number of categories -- using decision trees is not recommended for such datasets in general since the categorical features are favored over continuous features. Moreover, the user can use a combination of tricks (increasing bin size of the tree algorithms, use binary encoding for categorical features or use one-vs-all classification strategy) to avoid these constraints.
The new code is accompanied by unit tests and has also been tested on the iris and covtype datasets.
cc: mengxr, etrain, hirakendu, atalwalkar, srowen
Author: Manish Amde <manish9ue@gmail.com>
Author: manishamde <manish9ue@gmail.com>
Author: Evan Sparks <sparks@cs.berkeley.edu>
Closes#886 from manishamde/multiclass and squashes the following commits:
26f8acc [Manish Amde] another attempt at fixing mima
c5b2d04 [Manish Amde] more MIMA fixes
1ce7212 [Manish Amde] change problem filter for mima
10fdd82 [Manish Amde] fixing MIMA excludes
e1c970d [Manish Amde] merged master
abf2901 [Manish Amde] adding classes to MimaExcludes.scala
45e767a [Manish Amde] adding developer api annotation for overriden methods
c8428c4 [Manish Amde] fixing weird multiline bug
afced16 [Manish Amde] removed label weights support
2d85a48 [Manish Amde] minor: fixed scalastyle issues reprise
4e85f2c [Manish Amde] minor: fixed scalastyle issues
b2ae41f [Manish Amde] minor: scalastyle
e4c1321 [Manish Amde] using while loop for regression histograms
d75ac32 [Manish Amde] removed WeightedLabeledPoint from this PR
0fecd38 [Manish Amde] minor: add newline to EOF
2061cf5 [Manish Amde] merged from master
06b1690 [Manish Amde] fixed off-by-one error in bin to split conversion
9cc3e31 [Manish Amde] added implicit conversion import
5c1b2ca [Manish Amde] doc for PointConverter class
485eaae [Manish Amde] implicit conversion from LabeledPoint to WeightedLabeledPoint
3d7f911 [Manish Amde] updated doc
8e44ab8 [Manish Amde] updated doc
adc7315 [Manish Amde] support ordered categorical splits for multiclass classification
e3e8843 [Manish Amde] minor code formatting
23d4268 [Manish Amde] minor: another minor code style
34ee7b9 [Manish Amde] minor: code style
237762d [Manish Amde] renaming functions
12e6d0a [Manish Amde] minor: removing line in doc
9a90c93 [Manish Amde] Merge branch 'master' into multiclass
1892a2c [Manish Amde] tests and use multiclass binaggregate length when atleast one categorical feature is present
f5f6b83 [Manish Amde] multiclass for continous variables
8cfd3b6 [Manish Amde] working for categorical multiclass classification
828ff16 [Manish Amde] added categorical variable test
bce835f [Manish Amde] code cleanup
7e5f08c [Manish Amde] minor doc
1dd2735 [Manish Amde] bin search logic for multiclass
f16a9bb [Manish Amde] fixing while loop
d811425 [Manish Amde] multiclass bin aggregate logic
ab5cb21 [Manish Amde] multiclass logic
d8e4a11 [Manish Amde] sample weights
ed5a2df [Manish Amde] fixed classification requirements
d012be7 [Manish Amde] fixed while loop
18d2835 [Manish Amde] changing default values for num classes
6b912dc [Manish Amde] added numclasses to tree runner, predict logic for multiclass, add multiclass option to train
75f2bfc [Manish Amde] minor code style fix
e547151 [Manish Amde] minor modifications
34549d0 [Manish Amde] fixing error during merge
098e8c5 [Manish Amde] merged master
e006f9d [Manish Amde] changing variable names
5c78e1a [Manish Amde] added multiclass support
6c7af22 [Manish Amde] prepared for multiclass without breaking binary classification
46e06ee [Manish Amde] minor mods
3f85a17 [Manish Amde] tests for multiclass classification
4d5f70c [Manish Amde] added multiclass support for find splits bins
46f909c [Manish Amde] todo for multiclass support
455bea9 [Manish Amde] fixed tests
14aea48 [Manish Amde] changing instance format to weighted labeled point
a1a6e09 [Manish Amde] added weighted point class
968ca9d [Manish Amde] merged master
7fc9545 [Manish Amde] added docs
ce004a1 [Manish Amde] minor formatting
b27ad2c [Manish Amde] formatting
426bb28 [Manish Amde] programming guide blurb
8053fed [Manish Amde] more formatting
5eca9e4 [Manish Amde] grammar
4731cda [Manish Amde] formatting
5e82202 [Manish Amde] added documentation, fixed off by 1 error in max level calculation
cbd9f14 [Manish Amde] modified scala.math to math
dad9652 [Manish Amde] removed unused imports
e0426ee [Manish Amde] renamed parameter
718506b [Manish Amde] added unit test
1517155 [Manish Amde] updated documentation
9dbdabe [Manish Amde] merge from master
719d009 [Manish Amde] updating user documentation
fecf89a [manishamde] Merge pull request #6 from etrain/deep_tree
0287772 [Evan Sparks] Fixing scalastyle issue.
2f1e093 [Manish Amde] minor: added doc for maxMemory parameter
2f6072c [manishamde] Merge pull request #5 from etrain/deep_tree
abc5a23 [Evan Sparks] Parameterizing max memory.
50b143a [Manish Amde] adding support for very deep trees
Added check to LocalKMeans.scala: kMeansPlusPlus initialization to handle case with fewer distinct data points than clusters k. Added two related unit tests to KMeansSuite. (Re-submitting PR after tangling commits in PR 1407 https://github.com/apache/spark/pull/1407 )
Author: Joseph K. Bradley <joseph.kurata.bradley@gmail.com>
Closes#1468 from jkbradley/kmeans-fix and squashes the following commits:
4e9bd1e [Joseph K. Bradley] Updated PR per comments from mengxr
6c7a2ec [Joseph K. Bradley] Added check to LocalKMeans.scala: kMeansPlusPlus initialization to handle case with fewer distinct data points than clusters k. Added two related unit tests to KMeansSuite.
Instead of using prependOne currently in GeneralizedLinearAlgorithm, we would like to use appendBias for 1) keeping the indices of original training set unchanged by adding the intercept into the last element of vector and 2) using the same public API for consistently adding intercept.
Author: DB Tsai <dbtsai@alpinenow.com>
Closes#1410 from dbtsai/SPARK-2477_intercept_with_appendBias and squashes the following commits:
011432c [DB Tsai] From Alpine Data Labs
Apologies if there's an already-discussed reason I missed for why this doesn't make sense.
Author: Sandy Ryza <sandy@cloudera.com>
Closes#1389 from sryza/sandy-spark-2462 and squashes the following commits:
2e5e201 [Sandy Ryza] SPARK-2462. Make Vector.apply public.
After running some more tests on large matrix, found that the BV axpy (breeze/linalg/Vector.scala, axpy) is slower than the BSV axpy (breeze/linalg/operators/SparseVectorOps.scala, sv_dv_axpy), 8s v.s. 2s for each multiplication. The BV axpy operates on an iterator while BSV axpy directly operates on the underlying array. I think the overhead comes from creating the iterator (with a zip) and advancing the pointers.
Author: Li Pu <lpu@twitter.com>
Author: Xiangrui Meng <meng@databricks.com>
Author: Li Pu <li.pu@outlook.com>
Closes#1378 from vrilleup/master and squashes the following commits:
6fb01a3 [Li Pu] use specialized axpy in RowMatrix
5255f2a [Li Pu] Merge remote-tracking branch 'upstream/master'
7312ec1 [Li Pu] very minor comment fix
4c618e9 [Li Pu] Merge pull request #1 from mengxr/vrilleup-master
a461082 [Xiangrui Meng] make superscript show up correctly in doc
861ec48 [Xiangrui Meng] simplify axpy
62969fa [Xiangrui Meng] use BDV directly in symmetricEigs change the computation mode to local-svd, local-eigs, and dist-eigs update tests and docs
c273771 [Li Pu] automatically determine SVD compute mode and parameters
7148426 [Li Pu] improve RowMatrix multiply
5543cce [Li Pu] improve svd api
819824b [Li Pu] add flag for dense svd or sparse svd
eb15100 [Li Pu] fix binary compatibility
4c7aec3 [Li Pu] improve comments
e7850ed [Li Pu] use aggregate and axpy
827411b [Li Pu] fix EOF new line
9c80515 [Li Pu] use non-sparse implementation when k = n
fe983b0 [Li Pu] improve scala style
96d2ecb [Li Pu] improve eigenvalue sorting
e1db950 [Li Pu] SPARK-1782: svd for sparse matrix using ARPACK
It basically moved the private ColumnStatisticsAggregator class from RowMatrix to public available DeveloperApi with documentation and unitests.
Changes:
1) Moved the private implementation from org.apache.spark.mllib.linalg.ColumnStatisticsAggregator to org.apache.spark.mllib.stat.MultivariateOnlineSummarizer
2) When creating OnlineSummarizer object, the number of columns is not needed in the constructor. It's determined when users add the first sample.
3) Added the APIs documentation for MultivariateOnlineSummarizer.
4) Added the unittests for MultivariateOnlineSummarizer.
Author: DB Tsai <dbtsai@dbtsai.com>
Closes#955 from dbtsai/dbtsai-summarizer and squashes the following commits:
b13ac90 [DB Tsai] dbtsai-summarizer
copy ARPACK dsaupd/dseupd code from latest breeze
change RowMatrix to use sparse SVD
change tests for sparse SVD
All tests passed. I will run it against some large matrices.
Author: Li Pu <lpu@twitter.com>
Author: Xiangrui Meng <meng@databricks.com>
Author: Li Pu <li.pu@outlook.com>
Closes#964 from vrilleup/master and squashes the following commits:
7312ec1 [Li Pu] very minor comment fix
4c618e9 [Li Pu] Merge pull request #1 from mengxr/vrilleup-master
a461082 [Xiangrui Meng] make superscript show up correctly in doc
861ec48 [Xiangrui Meng] simplify axpy
62969fa [Xiangrui Meng] use BDV directly in symmetricEigs change the computation mode to local-svd, local-eigs, and dist-eigs update tests and docs
c273771 [Li Pu] automatically determine SVD compute mode and parameters
7148426 [Li Pu] improve RowMatrix multiply
5543cce [Li Pu] improve svd api
819824b [Li Pu] add flag for dense svd or sparse svd
eb15100 [Li Pu] fix binary compatibility
4c7aec3 [Li Pu] improve comments
e7850ed [Li Pu] use aggregate and axpy
827411b [Li Pu] fix EOF new line
9c80515 [Li Pu] use non-sparse implementation when k = n
fe983b0 [Li Pu] improve scala style
96d2ecb [Li Pu] improve eigenvalue sorting
e1db950 [Li Pu] SPARK-1782: svd for sparse matrix using ARPACK
Fixes test failures introduced by https://github.com/apache/spark/pull/1316.
For both the regression and classification cases,
val stats is the InformationGainStats for the best tree split.
stats.predict is the predicted value for the data, before the split is made.
Since 600 of the 1,000 values generated by DecisionTreeSuite.generateCategoricalDataPoints() are 1.0 and the rest 0.0, the regression tree and classification tree both correctly predict a value of 0.6 for this data now, and the assertions have been changed to reflect that.
Author: johnnywalleye <jsondag@gmail.com>
Closes#1343 from johnnywalleye/decision-tree-tests and squashes the following commits:
ef80603 [johnnywalleye] [SPARK-2417][MLlib] Fix DecisionTree tests
Hi, this pull fixes (what I believe to be) a bug in DecisionTree.scala.
In the extractLeftRightNodeAggregates function, the first set of rightNodeAgg values for Regression are set in line 792 as follows:
rightNodeAgg(featureIndex)(2 * (numBins - 2))
= binData(shift + (2 * numBins - 1)))
Then there is a loop that sets the rest of the values, as in line 809:
rightNodeAgg(featureIndex)(2 * (numBins - 2 - splitIndex)) =
binData(shift + (2 *(numBins - 2 - splitIndex))) +
rightNodeAgg(featureIndex)(2 * (numBins - 1 - splitIndex))
But since splitIndex starts at 1, this ends up skipping a set of binData values.
The changes here address this issue, for both the Regression and Classification cases.
Author: johnnywalleye <jsondag@gmail.com>
Closes#1316 from johnnywalleye/master and squashes the following commits:
73809da [johnnywalleye] fix bin offset in DecisionTree node aggregations
Just closing out this small JIRA, resolving with a comment change.
Author: Sean Owen <sowen@cloudera.com>
Closes#1171 from srowen/SPARK-1675 and squashes the following commits:
45ee9b7 [Sean Owen] Add simple note that data need not be centered for computePrincipalComponents
https://issues.apache.org/jira/browse/SPARK-2163
This pull request includes the change for **[SPARK-2163]**:
* Changed the convergence tolerance parameter from type `Int` to type `Double`.
* Added types for vars in `class LBFGS`, making the style consistent with `class GradientDescent`.
* Added associated test to check that optimizing via `class LBFGS` produces the same results as via calling `runLBFGS` from `object LBFGS`.
This is a very minor change but it will solve the problem in my implementation of a regression model for count data, where I make use of LBFGS for parameter estimation.
Author: Gang Bai <me@baigang.net>
Closes#1104 from BaiGang/fix_int_tol and squashes the following commits:
cecf02c [Gang Bai] Changed setConvergenceTol'' to specify tolerance with a parameter of type Double. For the reason and the problem caused by an Int parameter, please check https://issues.apache.org/jira/browse/SPARK-2163. Added a test in LBFGSSuite for validating that optimizing via class LBFGS produces the same results as calling runLBFGS from object LBFGS. Keep the indentations and styles correct.
in updateNumRows method in RowMatrix
Author: Doris Xin <doris.s.xin@gmail.com>
Closes#1125 from dorx/updateNumRows and squashes the following commits:
8564aef [Doris Xin] Squishing a typo bug before it causes real harm
The current implementation of ALS takes a single regularization parameter and apply it on both of the user factors and the product factors. This kind of regularization can be less effective while user number is significantly larger than the number of products (and vice versa). For example, if we have 10M users and 1K product, regularization on user factors will dominate. Following the discussion in [this thread](http://apache-spark-user-list.1001560.n3.nabble.com/possible-bug-in-Spark-s-ALS-implementation-tt2567.html#a2704), the implementation in this PR will regularize each factor vector by #ratings * lambda.
Author: Shuo Xiang <sxiang@twitter.com>
Closes#1026 from coderxiang/als-reg and squashes the following commits:
93dfdb4 [Shuo Xiang] Merge remote-tracking branch 'upstream/master' into als-reg
b98f19c [Shuo Xiang] merge latest master
52c7b58 [Shuo Xiang] Apply user-specific regularization instead of uniform regularization in Alternating Least Squares (ALS)
Some clean up work following #593.
1. Allow to set different number user blocks and number product blocks in `ALS`.
2. Update `MovieLensALS` to reflect the change.
Author: Tor Myklebust <tmyklebu@gmail.com>
Author: Xiangrui Meng <meng@databricks.com>
Closes#1014 from mengxr/SPARK-1672 and squashes the following commits:
0e910dd [Xiangrui Meng] change private[this] to private[recommendation]
36420c7 [Xiangrui Meng] set exclusion rules for ALS
9128b77 [Xiangrui Meng] Merge remote-tracking branch 'apache/master' into SPARK-1672
294efe9 [Xiangrui Meng] Merge remote-tracking branch 'apache/master' into SPARK-1672
9bab77b [Xiangrui Meng] clean up add numUserBlocks and numProductBlocks to MovieLensALS
84c8e8c [Xiangrui Meng] Merge branch 'master' into SPARK-1672
d17a8bf [Xiangrui Meng] merge master
a4925fd [Tor Myklebust] Style.
bd8a75c [Tor Myklebust] Merge branch 'master' of github.com:apache/spark into alsseppar
021f54b [Tor Myklebust] Separate user and product blocks.
dcf583a [Tor Myklebust] Remove the partitioner member variable; instead, thread that needle everywhere it needs to go.
23d6f91 [Tor Myklebust] Stop making the partitioner configurable.
495784f [Tor Myklebust] Merge branch 'master' of https://github.com/apache/spark
674933a [Tor Myklebust] Fix style.
40edc23 [Tor Myklebust] Fix missing space.
f841345 [Tor Myklebust] Fix daft bug creating 'pairs', also for -> foreach.
5ec9e6c [Tor Myklebust] Clean a couple of things up using 'map'.
36a0f43 [Tor Myklebust] Make the partitioner private.
d872b09 [Tor Myklebust] Add negative id ALS test.
df27697 [Tor Myklebust] Support custom partitioners. Currently we use the same partitioner for users and products.
c90b6d8 [Tor Myklebust] Scramble user and product ids before bucketing.
c774d7d [Tor Myklebust] Make the partitioner a member variable and use it instead of modding directly.
Author: witgo <witgo@qq.com>
Closes#1032 from witgo/ShouldMatchers and squashes the following commits:
7ebf34c [witgo] Resolve scalatest warnings during build
We should standardize the text format used to represent vectors and labeled points. The proposed formats are the following:
1. dense vector: `[v0,v1,..]`
2. sparse vector: `(size,[i0,i1],[v0,v1])`
3. labeled point: `(label,vector)`
where "(..)" indicates a tuple and "[...]" indicate an array. `loadLabeledPoints` is added to pyspark's `MLUtils`. I didn't add `loadVectors` to pyspark because `RDD.saveAsTextFile` cannot stringify dense vectors in the proposed format automatically.
`MLUtils#saveLabeledData` and `MLUtils#loadLabeledData` are deprecated. Users should use `RDD#saveAsTextFile` and `MLUtils#loadLabeledPoints` instead. In Scala, `MLUtils#loadLabeledPoints` is compatible with the format used by `MLUtils#loadLabeledData`.
CC: @mateiz, @srowen
Author: Xiangrui Meng <meng@databricks.com>
Closes#685 from mengxr/labeled-io and squashes the following commits:
2d1116a [Xiangrui Meng] make loadLabeledData/saveLabeledData deprecated since 1.0.1
297be75 [Xiangrui Meng] change LabeledPoint.parse to LabeledPointParser.parse to maintain binary compatibility
d6b1473 [Xiangrui Meng] Merge branch 'master' into labeled-io
56746ea [Xiangrui Meng] replace # by .
623a5f0 [Xiangrui Meng] merge master
f06d5ba [Xiangrui Meng] add docs and minor updates
640fe0c [Xiangrui Meng] throw SparkException
5bcfbc4 [Xiangrui Meng] update test to add scientific notations
e86bf38 [Xiangrui Meng] remove NumericTokenizer
050fca4 [Xiangrui Meng] use StringTokenizer
6155b75 [Xiangrui Meng] merge master
f644438 [Xiangrui Meng] remove parse methods based on eval from pyspark
a41675a [Xiangrui Meng] python loadLabeledPoint uses Scala's implementation
ce9a475 [Xiangrui Meng] add deserialize_labeled_point to pyspark with tests
e9fcd49 [Xiangrui Meng] add serializeLabeledPoint and tests
aea4ae3 [Xiangrui Meng] minor updates
810d6df [Xiangrui Meng] update tokenizer/parser implementation
7aac03a [Xiangrui Meng] remove Scala parsers
c1885c1 [Xiangrui Meng] add headers and minor changes
b0c50cb [Xiangrui Meng] add customized parser
d731817 [Xiangrui Meng] style update
63dc396 [Xiangrui Meng] add loadLabeledPoints to pyspark
ea122b5 [Xiangrui Meng] Merge branch 'master' into labeled-io
cd6c78f [Xiangrui Meng] add __str__ and parse to LabeledPoint
a7a178e [Xiangrui Meng] add stringify to pyspark's Vectors
5c2dbfa [Xiangrui Meng] add parse to pyspark's Vectors
7853f88 [Xiangrui Meng] update pyspark's SparseVector.__str__
e761d32 [Xiangrui Meng] make LabelPoint.parse compatible with the dense format used before v1.0 and deprecate loadLabeledData and saveLabeledData
9e63a02 [Xiangrui Meng] add loadVectors and loadLabeledPoints
19aa523 [Xiangrui Meng] update toString and add parsers for Vectors and LabeledPoint
This is very useful when debugging & fine tuning jobs with large data sets.
Author: Neville Li <neville@spotify.com>
Closes#966 from nevillelyh/master and squashes the following commits:
6747764 [Neville Li] [MLLIB] use string interpolation for RDD names
3b15d34 [Neville Li] [MLLIB] set RDD names in ALS
in RowMatrix.scala
Author: DB Tsai <dbtsai@dbtsai.com>
Closes#959 from dbtsai/dbtsai-typo and squashes the following commits:
fab0e0e [DB Tsai] Fixed typo
stop resetting spark.driver.port in unit tests (scala, java and python).
Author: Syed Hashmi <shashmi@cloudera.com>
Author: CodingCat <zhunansjtu@gmail.com>
Closes#943 from syedhashmi/master and squashes the following commits:
885f210 [Syed Hashmi] Removing unnecessary file (created by mergetool)
b8bd4b5 [Syed Hashmi] Merge remote-tracking branch 'upstream/master'
b895e59 [Syed Hashmi] Revert "[SPARK-1784] Add a new partitioner"
57b6587 [Syed Hashmi] Revert "[SPARK-1784] Add a balanced partitioner"
1574769 [Syed Hashmi] [SPARK-1942] Stop clearing spark.driver.port in unit tests
4354836 [Syed Hashmi] Revert "SPARK-1686: keep schedule() calling in the main thread"
fd36542 [Syed Hashmi] [SPARK-1784] Add a balanced partitioner
6668015 [CodingCat] SPARK-1686: keep schedule() calling in the main thread
4ca94cc [Syed Hashmi] [SPARK-1784] Add a new partitioner
This pull request includes a nonnegative least-squares solver (NNLS) tailored to the kinds of small-scale problems that come up when training matrix factorisation models by alternating nonnegative least-squares (ANNLS).
The method used for the NNLS subproblems is based on the classical method of projected gradients. There is a modification where, if the set of active constraints has not changed since the last iteration, a conjugate gradient step is considered and possibly rejected in favour of the gradient; this improves convergence once the optimal face has been located.
The NNLS solver is in `org.apache.spark.mllib.optimization.NNLSbyPCG`.
Author: Tor Myklebust <tmyklebu@gmail.com>
Closes#460 from tmyklebu/annls and squashes the following commits:
79bc4b5 [Tor Myklebust] Merge branch 'master' of https://github.com/apache/spark into annls
199b0bc [Tor Myklebust] Make the ctor private again and use the builder pattern.
7fbabf1 [Tor Myklebust] Cleanup matrix math in NNLSSuite.
65ef7f2 [Tor Myklebust] Make ALS's ctor public and remove a couple of "convenience" wrappers.
2d4f3cb [Tor Myklebust] Cleanup.
0cb4481 [Tor Myklebust] Drop the iteration limit from 40k to max(400,20n).
e2a01d1 [Tor Myklebust] Create a workspace object for NNLS to cut down on memory allocations.
b285106 [Tor Myklebust] Clean up NNLS test cases.
9c820b6 [Tor Myklebust] Tweak variable names.
8a1a436 [Tor Myklebust] Describe the problem and add a reference to Polyak's paper.
5345402 [Tor Myklebust] Style fixes that got eaten.
ac673bd [Tor Myklebust] More safeguards against numerical ridiculousness.
c288b6a [Tor Myklebust] Finish moving the NNLS solver.
9a82fa6 [Tor Myklebust] Fix scalastyle moanings.
33bf4f2 [Tor Myklebust] Fix missing space.
89ea0a8 [Tor Myklebust] Hack ALSSuite to support NNLS testing.
f5dbf4d [Tor Myklebust] Teach ALS how to use the NNLS solver.
6cb563c [Tor Myklebust] Tests for the nonnegative least squares solver.
a68ac10 [Tor Myklebust] A nonnegative least-squares solver.
JIRA: https://issues.apache.org/jira/browse/SPARK-1925
Author: zsxwing <zsxwing@gmail.com>
Closes#879 from zsxwing/SPARK-1925 and squashes the following commits:
5cf5a6d [zsxwing] SPARK-1925: Replace '&' with '&&'
the same reason as https://github.com/apache/spark/pull/588
Author: baishuo(白硕) <vc_java@hotmail.com>
Closes#815 from baishuo/master and squashes the following commits:
6876c1e [baishuo(白硕)] Update LBFGSSuite.scala
`model.predict` returns a RDD of Scala primitive type (Int/Double), which is recognized as Object in Java. Adding predict(JavaRDD) could make life easier for Java users.
Added tests for KMeans, LinearRegression, and NaiveBayes.
Will update examples after https://github.com/apache/spark/pull/653 gets merged.
cc: @srowen
Author: Xiangrui Meng <meng@databricks.com>
Closes#670 from mengxr/predict-javardd and squashes the following commits:
b77ccd8 [Xiangrui Meng] Merge branch 'master' into predict-javardd
43caac9 [Xiangrui Meng] add predict(JavaRDD) to RegressionModel, ClassificationModel, and KMeans
This is a few changes based on the original patch by @scrapcodes.
Author: Prashant Sharma <prashant.s@imaginea.com>
Author: Patrick Wendell <pwendell@gmail.com>
Closes#785 from pwendell/package-docs and squashes the following commits:
c32b731 [Patrick Wendell] Changes based on Prashant's patch
c0463d3 [Prashant Sharma] added eof new line
ce8bf73 [Prashant Sharma] Added eof new line to all files.
4c35f2e [Prashant Sharma] SPARK-1563 Add package-info.java and package.scala files for all packages that appear in docs
It doesn't affect existing code because only `alpha = 1.0` is used in the code.
Author: Xiangrui Meng <meng@databricks.com>
Closes#778 from mengxr/mllib-dspr-fix and squashes the following commits:
a37402e [Xiangrui Meng] use alpha in dense dspr
Summary:
https://issues.apache.org/jira/browse/SPARK-1791
Simple fix, and backward compatible, since
- anyone who set the threshold was getting completely wrong answers.
- anyone who did not set the threshold had the default 0.0 value for the threshold anyway.
Test Plan:
Unit test added that is verified to fail under the old implementation,
and pass under the new implementation.
Reviewers:
CC:
Author: Andrew Tulloch <andrew@tullo.ch>
Closes#725 from ajtulloch/SPARK-1791-SVM and squashes the following commits:
770f55d [Andrew Tulloch] SPARK-1791 - SVM implementation does not use threshold parameter
Three issues related to temp files that tests generate – these should be touched up for hygiene but are not urgent.
Modules have a log4j.properties which directs the unit-test.log output file to a directory like `[module]/target/unit-test.log`. But this ends up creating `[module]/[module]/target/unit-test.log` instead of former.
The `work/` directory is not deleted by "mvn clean", in the parent and in modules. Neither is the `checkpoint/` directory created under the various external modules.
Many tests create a temp directory, which is not usually deleted. This can be largely resolved by calling `deleteOnExit()` at creation and trying to call `Utils.deleteRecursively` consistently to clean up, sometimes in an `@After` method.
_If anyone seconds the motion, I can create a more significant change that introduces a new test trait along the lines of `LocalSparkContext`, which provides management of temp directories for subclasses to take advantage of._
Author: Sean Owen <sowen@cloudera.com>
Closes#732 from srowen/SPARK-1798 and squashes the following commits:
5af578e [Sean Owen] Try to consistently delete test temp dirs and files, and set deleteOnExit() for each
b21b356 [Sean Owen] Remove work/ and checkpoint/ dirs with mvn clean
bdd0f41 [Sean Owen] Remove duplicate module dir in log4j.properties output path for tests
Fixed a small bug caused by the inconsistency of index/data array size and vector length.
Author: Funes <tianshaocun@gmail.com>
Author: funes <tianshaocun@gmail.com>
Closes#661 from funes/bugfix and squashes the following commits:
edb2b9d [funes] remove unused import
75dced3 [Funes] update test case
d129a66 [Funes] Add test for sparse breeze by vector builder
64e7198 [Funes] Copy data only when necessary
b85806c [Funes] Bug fix of sparse vector conversion
Getting the lossHistory from Breeze's API which already excludes the rejection steps in line search. Also, remove the miniBatch in LBFGS since those quasi-Newton methods approximate the inverse of Hessian. It doesn't make sense if the gradients are computed from a varying objective.
Author: DB Tsai <dbtsai@alpinenow.com>
Closes#582 from dbtsai/dbtsai-lbfgs-bug and squashes the following commits:
9cc6cf9 [DB Tsai] Removed the miniBatch in LBFGS.
1ba6a33 [DB Tsai] Formatting the code.
d72c679 [DB Tsai] Using Breeze's states to get the loss.
@etrain and I came with a PR for arbitrarily deep decision trees at the cost of multiple passes over the data at deep tree levels.
To summarize:
1) We take a parameter that indicates the amount of memory users want to reserve for computation on each worker (and 2x that at the driver).
2) Using that information, we calculate two things - the maximum depth to which we train as usual (which is, implicitly, the maximum number of nodes we want to train in parallel), and the size of the groups we should use in the case where we exceed this depth.
cc: @atalwalkar, @hirakendu, @mengxr
Author: Manish Amde <manish9ue@gmail.com>
Author: manishamde <manish9ue@gmail.com>
Author: Evan Sparks <sparks@cs.berkeley.edu>
Closes#475 from manishamde/deep_tree and squashes the following commits:
968ca9d [Manish Amde] merged master
7fc9545 [Manish Amde] added docs
ce004a1 [Manish Amde] minor formatting
b27ad2c [Manish Amde] formatting
426bb28 [Manish Amde] programming guide blurb
8053fed [Manish Amde] more formatting
5eca9e4 [Manish Amde] grammar
4731cda [Manish Amde] formatting
5e82202 [Manish Amde] added documentation, fixed off by 1 error in max level calculation
cbd9f14 [Manish Amde] modified scala.math to math
dad9652 [Manish Amde] removed unused imports
e0426ee [Manish Amde] renamed parameter
718506b [Manish Amde] added unit test
1517155 [Manish Amde] updated documentation
9dbdabe [Manish Amde] merge from master
719d009 [Manish Amde] updating user documentation
fecf89a [manishamde] Merge pull request #6 from etrain/deep_tree
0287772 [Evan Sparks] Fixing scalastyle issue.
2f1e093 [Manish Amde] minor: added doc for maxMemory parameter
2f6072c [manishamde] Merge pull request #5 from etrain/deep_tree
abc5a23 [Evan Sparks] Parameterizing max memory.
50b143a [Manish Amde] adding support for very deep trees
use more faster way to construct an array
Author: baishuo(白硕) <vc_java@hotmail.com>
Closes#588 from baishuo/master and squashes the following commits:
45b95fb [baishuo(白硕)] Update GradientDescentSuite.scala
c03b61c [baishuo(白硕)] Update GradientDescentSuite.scala
b666d27 [baishuo(白硕)] Update GradientDescentSuite.scala
Final pass before the v1.0 release.
* Remove `VectorRDDs`
* Move `BinaryClassificationMetrics` from `evaluation.binary` to `evaluation`
* Change default value of `addIntercept` to false and allow to add intercept in Ridge and Lasso.
* Clean `DecisionTree` package doc and test suite.
* Mark model constructors `private[spark]`
* Rename `loadLibSVMData` to `loadLibSVMFile` and hide `LabelParser` from users.
* Add `saveAsLibSVMFile`.
* Add `appendBias` to `MLUtils`.
Author: Xiangrui Meng <meng@databricks.com>
Closes#524 from mengxr/mllib-cleaning and squashes the following commits:
295dc8b [Xiangrui Meng] update loadLibSVMFile doc
1977ac1 [Xiangrui Meng] fix doc of appendBias
649fcf0 [Xiangrui Meng] rename loadLibSVMData to loadLibSVMFile; hide LabelParser from user APIs
54b812c [Xiangrui Meng] add appendBias
a71e7d0 [Xiangrui Meng] add saveAsLibSVMFile
d976295 [Xiangrui Meng] Merge branch 'master' into mllib-cleaning
b7e5cec [Xiangrui Meng] remove some experimental annotations and make model constructors private[mllib]
9b02b93 [Xiangrui Meng] minor code style update
a593ddc [Xiangrui Meng] fix python tests
fc28c18 [Xiangrui Meng] mark more classes experimental
f6cbbff [Xiangrui Meng] fix Java tests
0af70b0 [Xiangrui Meng] minor
6e139ef [Xiangrui Meng] Merge branch 'master' into mllib-cleaning
94e6dce [Xiangrui Meng] move BinaryLabelCounter and BinaryConfusionMatrixImpl to evaluation.binary
df34907 [Xiangrui Meng] clean DecisionTreeSuite to use LocalSparkContext
c81807f [Xiangrui Meng] set the default value of AddIntercept to false
03389c0 [Xiangrui Meng] allow to add intercept in Ridge and Lasso
c66c56f [Xiangrui Meng] move tree md to package object doc
a2695df [Xiangrui Meng] update guide for BinaryClassificationMetrics
9194f4c [Xiangrui Meng] move BinaryClassificationMetrics one level up
1c1a0e3 [Xiangrui Meng] remove VectorRDDs because it only contains one function that is not necessary for us to maintain
This change replaces some Scala `for` and `foreach` constructs with `while` constructs. There may be a slight performance gain on the order of 1-2% when training an ALS model.
I trained an ALS model on the Movielens 10M-rating dataset repeatedly both with and without these changes. All 7 runs in both columns were done in a Scala `for` loop like this:
for (iter <- 0 to 10) {
val before = System.currentTimeMillis()
val model = ALS.train(rats, 20, 10)
val after = System.currentTimeMillis()
println("%d ms".format(after-before))
println("rmse %g".format(computeRmse(model, rats, numRatings)))
}
The timings were done on a multiuser machine, and I stopped one set of timings after 7 had been completed. It would be nice if somebody with dedicated hardware could confirm my timings.
After Before
121980 ms 122041 ms
117069 ms 117127 ms
115332 ms 117523 ms
115381 ms 117402 ms
114635 ms 116550 ms
114140 ms 114076 ms
112993 ms 117200 ms
Ratios are about 1.0005, 1.0005, 1.019, 1.0175, 1.01671, 0.99944, and 1.03723. I therefore suspect these changes make for a slight performance gain on the order of 1-2%.
Author: Tor Myklebust <tmyklebu@gmail.com>
Closes#568 from tmyklebu/alsopt and squashes the following commits:
5ded80f [Tor Myklebust] Fix style.
79595ff [Tor Myklebust] Fix style error.
4ef0313 [Tor Myklebust] Merge branch 'master' of github.com:apache/spark into alsopt
114fb74 [Tor Myklebust] Turn some 'for' loops into 'while' loops.
dcf583a [Tor Myklebust] Remove the partitioner member variable; instead, thread that needle everywhere it needs to go.
23d6f91 [Tor Myklebust] Stop making the partitioner configurable.
495784f [Tor Myklebust] Merge branch 'master' of https://github.com/apache/spark
674933a [Tor Myklebust] Fix style.
40edc23 [Tor Myklebust] Fix missing space.
f841345 [Tor Myklebust] Fix daft bug creating 'pairs', also for -> foreach.
5ec9e6c [Tor Myklebust] Clean a couple of things up using 'map'.
36a0f43 [Tor Myklebust] Make the partitioner private.
d872b09 [Tor Myklebust] Add negative id ALS test.
df27697 [Tor Myklebust] Support custom partitioners. Currently we use the same partitioner for users and products.
c90b6d8 [Tor Myklebust] Scramble user and product ids before bucketing.
c774d7d [Tor Myklebust] Make the partitioner a member variable and use it instead of modding directly.
* `NaiveBayes` -> `SparseNaiveBayes`
* `KMeans` -> `DenseKMeans`
* `SVMWithSGD` and `LogisticRegerssionWithSGD` -> `BinaryClassification`
* `ALS` -> `MovieLensALS`
* `LinearRegressionWithSGD`, `LassoWithSGD`, and `RidgeRegressionWithSGD` -> `LinearRegression`
* `DecisionTree` -> `DecisionTreeRunner`
`scopt` is used for parsing command-line parameters. `scopt` has MIT license and it only depends on `scala-library`.
Example help message:
~~~
BinaryClassification: an example app for binary classification.
Usage: BinaryClassification [options] <input>
--numIterations <value>
number of iterations
--stepSize <value>
initial step size, default: 1.0
--algorithm <value>
algorithm (SVM,LR), default: LR
--regType <value>
regularization type (L1,L2), default: L2
--regParam <value>
regularization parameter, default: 0.1
<input>
input paths to labeled examples in LIBSVM format
~~~
Author: Xiangrui Meng <meng@databricks.com>
Closes#584 from mengxr/mllib-main and squashes the following commits:
7b58c60 [Xiangrui Meng] minor
6e35d7e [Xiangrui Meng] make imports explicit and fix code style
c6178c9 [Xiangrui Meng] update TS PCA/SVD to use new spark-submit
6acff75 [Xiangrui Meng] use scopt for DecisionTreeRunner
be86069 [Xiangrui Meng] use main instead of extending App
b3edf68 [Xiangrui Meng] move DecisionTree's main method to examples
8bfaa5a [Xiangrui Meng] change NaiveBayesParams to Params
fe23dcb [Xiangrui Meng] remove main from KMeans and add DenseKMeans as an example
67f4448 [Xiangrui Meng] remove main methods from linear regression algorithms and add LinearRegression example
b066bbc [Xiangrui Meng] remove main from ALS and add MovieLensALS example
b040f3b [Xiangrui Meng] change BinaryClassificationParams to Params
577945b [Xiangrui Meng] remove unused imports from NB
3d299bc [Xiangrui Meng] remove main from LR/SVM and add an example app for binary classification
f70878e [Xiangrui Meng] remove main from NaiveBayes and add an example NaiveBayes app
01ec2cd [Xiangrui Meng] Merge branch 'master' into mllib-main
9420692 [Xiangrui Meng] add scopt to examples dependencies
Author: Sandeep <sandeep@techaddict.me>
Closes#503 from techaddict/fix-79 and squashes the following commits:
e3f6746 [Sandeep] Style changes
07a4f6b [Sandeep] for loop to While loop
0a6d8e9 [Sandeep] Breakable for loop to While loop
ALS was using HashPartitioner and explicit uses of `%` together. Further, the naked use of `%` meant that, if the number of partitions corresponded with the stride of arithmetic progressions appearing in user and product ids, users and products could be mapped into buckets in an unfair or unwise way.
This pull request:
1) Makes the Partitioner an instance variable of ALS.
2) Replaces the direct uses of `%` with calls to a Partitioner.
3) Defines an anonymous Partitioner that scrambles the bits of the object's hashCode before reducing to the number of present buckets.
This pull request does not make the partitioner user-configurable.
I'm not all that happy about the way I did (1). It introduces an icky lifetime issue and dances around it by nulling something. However, I don't know a better way to make the partitioner visible everywhere it needs to be visible.
Author: Tor Myklebust <tmyklebu@gmail.com>
Closes#407 from tmyklebu/master and squashes the following commits:
dcf583a [Tor Myklebust] Remove the partitioner member variable; instead, thread that needle everywhere it needs to go.
23d6f91 [Tor Myklebust] Stop making the partitioner configurable.
495784f [Tor Myklebust] Merge branch 'master' of https://github.com/apache/spark
674933a [Tor Myklebust] Fix style.
40edc23 [Tor Myklebust] Fix missing space.
f841345 [Tor Myklebust] Fix daft bug creating 'pairs', also for -> foreach.
5ec9e6c [Tor Myklebust] Clean a couple of things up using 'map'.
36a0f43 [Tor Myklebust] Make the partitioner private.
d872b09 [Tor Myklebust] Add negative id ALS test.
df27697 [Tor Myklebust] Support custom partitioners. Currently we use the same partitioner for users and products.
c90b6d8 [Tor Myklebust] Scramble user and product ids before bucketing.
c774d7d [Tor Myklebust] Make the partitioner a member variable and use it instead of modding directly.
... so that we don't follow an unspoken set of forbidden rules for adding **@AlphaComponent**, **@DeveloperApi**, and **@Experimental** annotations in the code.
In addition, this PR
(1) removes unnecessary `:: * ::` tags,
(2) adds missing `:: * ::` tags, and
(3) removes annotations for internal APIs.
Author: Andrew Or <andrewor14@gmail.com>
Closes#470 from andrewor14/annotations-fix and squashes the following commits:
92a7f42 [Andrew Or] Document + fix annotation usages
`new DoubleMatrix(double[])` creates a garbage `double[]` of the same length as its argument and immediately throws it away. This pull request avoids that constructor in the ALS code.
Author: Tor Myklebust <tmyklebu@gmail.com>
Closes#442 from tmyklebu/foo2 and squashes the following commits:
2784fc5 [Tor Myklebust] Mention that this is probably fixed as of jblas 1.2.4; repunctuate.
a09904f [Tor Myklebust] Helper function for wrapping Array[Double]'s with DoubleMatrix's.
Per discussion, this is my suggestion to make ALS Rating, ClassificationModel, RegressionModel experimental for now, to reserve the right to possibly change after 1.0. See what you think of this much.
Author: Sean Owen <sowen@cloudera.com>
Closes#372 from srowen/SPARK-1357Addendum and squashes the following commits:
17cf1ea [Sean Owen] Remove (another) blank line after ":: Experimental ::"
6800e4c [Sean Owen] Remove blank line after ":: Experimental ::"
b3a88d2 [Sean Owen] Make ALS Rating, ClassificationModel, RegressionModel experimental for now, to reserve the right to possibly change after 1.0
https://issues.apache.org/jira/browse/SPARK-1483
From the original JIRA: " The parameter name is part of the public API in Scala and Python, since you can pass named parameters to a method, so we should name it to this more descriptive term. Everywhere else we refer to "splits" as partitions." - @mateiz
Author: CodingCat <zhunansjtu@gmail.com>
Closes#430 from CodingCat/SPARK-1483 and squashes the following commits:
4b60541 [CodingCat] deprecate defaultMinSplits
ba2c663 [CodingCat] Rename minSplits to minPartitions in public APIs
Author: Holden Karau <holden@pigscanfly.ca>
Closes#18 from holdenk/addkfoldcrossvalidation and squashes the following commits:
208db9b [Holden Karau] Fix a bad space
e84f2fc [Holden Karau] Fix the test, we should be looking at the second element instead
6ddbf05 [Holden Karau] swap training and validation order
7157ae9 [Holden Karau] CR feedback
90896c7 [Holden Karau] New line
150889c [Holden Karau] Fix up error messages in the MLUtilsSuite
2cb90b3 [Holden Karau] Fix the names in kFold
c702a96 [Holden Karau] Fix imports in MLUtils
e187e35 [Holden Karau] Move { up to same line as whenExecuting(random) in RandomSamplerSuite.scala
c5b723f [Holden Karau] clean up
7ebe4d5 [Holden Karau] CR feedback, remove unecessary learners (came back during merge mistake) and insert an empty line
bb5fa56 [Holden Karau] extra line sadness
163c5b1 [Holden Karau] code review feedback 1.to -> 1 to and folds -> numFolds
5a33f1d [Holden Karau] Code review follow up.
e8741a7 [Holden Karau] CR feedback
b78804e [Holden Karau] Remove cross validation [TODO in another pull request]
91eae64 [Holden Karau] Consolidate things in mlutils
264502a [Holden Karau] Add a test for the bug that was found with BernoulliSampler not copying the complement param
dd0b737 [Holden Karau] Wrap long lines (oops)
c0b7fa4 [Holden Karau] Switch FoldedRDD to use BernoulliSampler and PartitionwiseSampledRDD
08f8e4d [Holden Karau] Fix BernoulliSampler to respect complement
a751ec6 [Holden Karau] Add k-fold cross validation to MLLib
This PR adds a SparseVector class in PySpark and updates all the regression, classification and clustering algorithms and models to support sparse data, similar to MLlib. I chose to add this class because SciPy is quite difficult to install in many environments (more so than NumPy), but I plan to add support for SciPy sparse vectors later too, and make the methods work transparently on objects of either type.
On the Scala side, we keep Python sparse vectors sparse and pass them to MLlib. We always return dense vectors from our models.
Some to-do items left:
- [x] Support SciPy's scipy.sparse matrix objects when SciPy is available. We can easily add a function to convert these to our own SparseVector.
- [x] MLlib currently uses a vector with one extra column on the left to represent what we call LabeledPoint in Scala. Do we really want this? It may get annoying once you deal with sparse data since you must add/subtract 1 to each feature index when training. We can remove this API in 1.0 and use tuples for labeling.
- [x] Explain how to use these in the Python MLlib docs.
CC @mengxr, @joshrosen
Author: Matei Zaharia <matei@databricks.com>
Closes#341 from mateiz/py-ml-update and squashes the following commits:
d52e763 [Matei Zaharia] Remove no-longer-needed slice code and handle review comments
ea5a25a [Matei Zaharia] Fix remaining uses of copyto() after merge
b9f97a3 [Matei Zaharia] Fix test
1e1bd0f [Matei Zaharia] Add MLlib logistic regression example in Python
88bc01f [Matei Zaharia] Clean up inheritance of LinearModel in Python, and expose its parametrs
37ab747 [Matei Zaharia] Fix some examples and docs due to changes in MLlib API
da0f27e [Matei Zaharia] Added a MLlib K-means example and updated docs to discuss sparse data
c48e85a [Matei Zaharia] Added some tests for passing lists as input, and added mllib/tests.py to run-tests script.
a07ba10 [Matei Zaharia] Fix some typos and calculation of initial weights
74eefe7 [Matei Zaharia] Added LabeledPoint class in Python
889dde8 [Matei Zaharia] Support scipy.sparse matrices in all our algorithms and models
ab244d1 [Matei Zaharia] Allow SparseVectors to be initialized using a dict
a5d6426 [Matei Zaharia] Add linalg.py to run-tests script
0e7a3d8 [Matei Zaharia] Keep vectors sparse in Java when reading LabeledPoints
eaee759 [Matei Zaharia] Update regression, classification and clustering models for sparse data
2abbb44 [Matei Zaharia] Further work to get linear models working with sparse data
154f45d [Matei Zaharia] Update docs, name some magic values
881fef7 [Matei Zaharia] Added a sparse vector in Python and made Java-Python format more compact
This PR uses Breeze's L-BFGS implement, and Breeze dependency has already been introduced by Xiangrui's sparse input format work in SPARK-1212. Nice work, @mengxr !
When use with regularized updater, we need compute the regVal and regGradient (the gradient of regularized part in the cost function), and in the currently updater design, we can compute those two values by the following way.
Let's review how updater works when returning newWeights given the input parameters.
w' = w - thisIterStepSize * (gradient + regGradient(w)) Note that regGradient is function of w!
If we set gradient = 0, thisIterStepSize = 1, then
regGradient(w) = w - w'
As a result, for regVal, it can be computed by
val regVal = updater.compute(
weights,
new DoubleMatrix(initialWeights.length, 1), 0, 1, regParam)._2
and for regGradient, it can be obtained by
val regGradient = weights.sub(
updater.compute(weights, new DoubleMatrix(initialWeights.length, 1), 1, 1, regParam)._1)
The PR includes the tests which compare the result with SGD with/without regularization.
We did a comparison between LBFGS and SGD, and often we saw 10x less
steps in LBFGS while the cost of per step is the same (just computing
the gradient).
The following is the paper by Prof. Ng at Stanford comparing different
optimizers including LBFGS and SGD. They use them in the context of
deep learning, but worth as reference.
http://cs.stanford.edu/~jngiam/papers/LeNgiamCoatesLahiriProchnowNg2011.pdf
Author: DB Tsai <dbtsai@alpinenow.com>
Closes#353 from dbtsai/dbtsai-LBFGS and squashes the following commits:
984b18e [DB Tsai] L-BFGS Optimizer based on Breeze's implementation. Also fixed indentation issue in GradientDescent optimizer.
For your consideration: scalac currently notes a number of feature warnings during compilation:
```
[warn] there were 65 feature warning(s); re-run with -feature for details
```
Warnings are like:
```
[warn] /Users/srowen/Documents/spark/core/src/main/scala/org/apache/spark/SparkContext.scala:1261: implicit conversion method rddToPairRDDFunctions should be enabled
[warn] by making the implicit value scala.language.implicitConversions visible.
[warn] This can be achieved by adding the import clause 'import scala.language.implicitConversions'
[warn] or by setting the compiler option -language:implicitConversions.
[warn] See the Scala docs for value scala.language.implicitConversions for a discussion
[warn] why the feature should be explicitly enabled.
[warn] implicit def rddToPairRDDFunctions[K: ClassTag, V: ClassTag](rdd: RDD[(K, V)]) =
[warn] ^
```
scalac is suggesting that it's just best practice to explicitly enable certain language features by importing them where used.
This PR simply adds the imports it suggests (and squashes one other Java warning along the way). This leaves just deprecation warnings in the build.
Author: Sean Owen <sowen@cloudera.com>
Closes#404 from srowen/SPARK-1488 and squashes the following commits:
8598980 [Sean Owen] Quiet scalac warnings about language features by explicitly importing language features.
39bc831 [Sean Owen] Enable -feature in scalac to emit language feature warnings
As with the new vector system in MLlib, we find that it is good to add some new APIs to precess the `RDD[Vector]`. Beside, the former implementation of `computeStat` is not stable which could loss precision, and has the possibility to cause `Nan` in scientific computing, just as said in the [SPARK-1328](https://spark-project.atlassian.net/browse/SPARK-1328).
APIs contain:
* rowMeans(): RDD[Double]
* rowNorm2(): RDD[Double]
* rowSDs(): RDD[Double]
* colMeans(): Vector
* colMeans(size: Int): Vector
* colNorm2(): Vector
* colNorm2(size: Int): Vector
* colSDs(): Vector
* colSDs(size: Int): Vector
* maxOption((Vector, Vector) => Boolean): Option[Vector]
* minOption((Vector, Vector) => Boolean): Option[Vector]
* rowShrink(): RDD[Vector]
* colShrink(): RDD[Vector]
This is working in process now, and some more APIs will add to `LabeledPoint`. Moreover, the implicit declaration will move from `MLUtils` to `MLContext` later.
Author: Xusen Yin <yinxusen@gmail.com>
Author: Xiangrui Meng <meng@databricks.com>
Closes#268 from yinxusen/vector-statistics and squashes the following commits:
d61363f [Xusen Yin] rebase to latest master
16ae684 [Xusen Yin] fix minor error and remove useless method
10cf5d3 [Xusen Yin] refine some return type
b064714 [Xusen Yin] remove computeStat in MLUtils
cbbefdb [Xiangrui Meng] update multivariate statistical summary interface and clean tests
4eaf28a [Xusen Yin] merge VectorRDDStatistics into RowMatrix
48ee053 [Xusen Yin] fix minor error
e624f93 [Xusen Yin] fix scala style error
1fba230 [Xusen Yin] merge while loop together
69e1f37 [Xusen Yin] remove lazy eval, and minor memory footprint
548e9de [Xusen Yin] minor revision
86522c4 [Xusen Yin] add comments on functions
dc77e38 [Xusen Yin] test sparse vector RDD
18cf072 [Xusen Yin] change def to lazy val to make sure that the computations in function be evaluated only once
f7a3ca2 [Xusen Yin] fix the corner case of maxmin
967d041 [Xusen Yin] full revision with Aggregator class
138300c [Xusen Yin] add new Aggregator class
1376ff4 [Xusen Yin] rename variables and adjust code
4a5c38d [Xusen Yin] add scala doc, refine code and comments
036b7a5 [Xusen Yin] fix the bug of Nan occur
f6e8e9a [Xusen Yin] add sparse vectors test
4cfbadf [Xusen Yin] fix bug of min max
4e4fbd1 [Xusen Yin] separate seqop and combop out as independent functions
a6d5a2e [Xusen Yin] rewrite for only computing non-zero elements
3980287 [Xusen Yin] rename variables
62a2c3e [Xusen Yin] use axpy and in-place if possible
9a75ebd [Xusen Yin] add case class to wrap return values
d816ac7 [Xusen Yin] remove useless APIs
c4651bb [Xusen Yin] remove row-wise APIs and refine code
1338ea1 [Xusen Yin] all-in-one version test passed
cc65810 [Xusen Yin] add parallel mean and variance
9af2e95 [Xusen Yin] refine the code style
ad6c82d [Xusen Yin] add shrink test
e09d5d2 [Xusen Yin] add scala docs and refine shrink method
8ef3377 [Xusen Yin] pass all tests
28cf060 [Xusen Yin] fix error of column means
54b19ab [Xusen Yin] add new API to shrink RDD[Vector]
8c6c0e1 [Xusen Yin] add basic statistics
This PR implements a generic version of `AreaUnderCurve` using the `RDD.sliding` implementation from https://github.com/apache/spark/pull/136 . It also contains refactoring of https://github.com/apache/spark/pull/160 for binary classification evaluation.
Author: Xiangrui Meng <meng@databricks.com>
Closes#364 from mengxr/auc and squashes the following commits:
a05941d [Xiangrui Meng] replace TP/FP/TN/FN by their full names
3f42e98 [Xiangrui Meng] add (0, 0), (1, 1) to roc, and (0, 1) to pr
fb4b6d2 [Xiangrui Meng] rename Evaluator to Metrics and add more metrics
b1b7dab [Xiangrui Meng] fix code styles
9dc3518 [Xiangrui Meng] add tests for BinaryClassificationEvaluator
ca31da5 [Xiangrui Meng] remove PredictionAndResponse
3d71525 [Xiangrui Meng] move binary evalution classes to evaluation.binary
8f78958 [Xiangrui Meng] add PredictionAndResponse
dda82d5 [Xiangrui Meng] add confusion matrix
aa7e278 [Xiangrui Meng] add initial version of binary classification evaluator
221ebce [Xiangrui Meng] add a new test to sliding
a920865 [Xiangrui Meng] Merge branch 'sliding' into auc
a9b250a [Xiangrui Meng] move sliding to mllib
cab9a52 [Xiangrui Meng] use last for the last element
db6cb30 [Xiangrui Meng] remove unnecessary toSeq
9916202 [Xiangrui Meng] change RDD.sliding return type to RDD[Seq[T]]
284d991 [Xiangrui Meng] change SlidedRDD to SlidingRDD
c1c6c22 [Xiangrui Meng] add AreaUnderCurve
65461b2 [Xiangrui Meng] Merge branch 'sliding' into auc
5ee6001 [Xiangrui Meng] add TODO
d2a600d [Xiangrui Meng] add sliding to rdd
stack these together in a commit else they show up chunk by chunk in different commits.
Author: Sandeep <sandeep@techaddict.me>
Closes#380 from techaddict/white_space and squashes the following commits:
b58f294 [Sandeep] Remove Unnecessary Whitespace's
Remove empty line after :: DeveloperApi/Experimental :: in comments to make the original doc show up in the preview of the generated html docs. Thanks @andrewor14 !
Author: Xiangrui Meng <meng@databricks.com>
Closes#373 from mengxr/api and squashes the following commits:
9c35bdc [Xiangrui Meng] remove the empty line after :: DeveloperApi/Experimental ::
Annotate developer and experimental APIs in MLlib.
Author: Xiangrui Meng <meng@databricks.com>
Closes#298 from mengxr/api and squashes the following commits:
13390e8 [Xiangrui Meng] Merge branch 'master' into api
dc4cbb3 [Xiangrui Meng] mark distribute matrices experimental
6b9f8e2 [Xiangrui Meng] add Experimental annotation
8773d0d [Xiangrui Meng] add DeveloperApi annotation
da31733 [Xiangrui Meng] update developer and experimental tags
555e0fe [Xiangrui Meng] Merge branch 'master' into api
ef1a717 [Xiangrui Meng] mark some constructors private add default parameters to JavaDoc
00ffbcc [Xiangrui Meng] update tree API annotation
0b674fa [Xiangrui Meng] mark decision tree APIs
86b9e34 [Xiangrui Meng] one pass over APIs of GLMs, NaiveBayes, and ALS
f21d862 [Xiangrui Meng] Merge branch 'master' into api
2b133d6 [Xiangrui Meng] intial annotation of developer and experimental apis
This is to refactor interfaces for matrices backed by RDDs. It would be better if we have a clear separation of local matrices and those backed by RDDs. Right now, we have
1. `org.apache.spark.mllib.linalg.SparseMatrix`, which is a wrapper over an RDD of matrix entries, i.e., coordinate list format.
2. `org.apache.spark.mllib.linalg.TallSkinnyDenseMatrix`, which is a wrapper over RDD[Array[Double]], i.e. row-oriented format.
We will see naming collision when we introduce local `SparseMatrix`, and the name `TallSkinnyDenseMatrix` is not exact if we switch to `RDD[Vector]` from `RDD[Array[Double]]`. It would be better to have "RDD" in the class name to suggest that operations may trigger jobs.
The proposed names are (all under `org.apache.spark.mllib.linalg.rdd`):
1. `RDDMatrix`: trait for matrices backed by one or more RDDs
2. `CoordinateRDDMatrix`: wrapper of `RDD[(Long, Long, Double)]`
3. `RowRDDMatrix`: wrapper of `RDD[Vector]` whose rows do not have special ordering
4. `IndexedRowRDDMatrix`: wrapper of `RDD[(Long, Vector)]` whose rows are associated with indices
The current code also introduces local matrices.
Author: Xiangrui Meng <meng@databricks.com>
Closes#296 from mengxr/mat and squashes the following commits:
24d8294 [Xiangrui Meng] fix for groupBy returning Iterable
bfc2b26 [Xiangrui Meng] merge master
8e4f1f5 [Xiangrui Meng] Merge branch 'master' into mat
0135193 [Xiangrui Meng] address Reza's comments
03cd7e1 [Xiangrui Meng] add pca/gram to IndexedRowMatrix add toBreeze to DistributedMatrix for test simplify tests
b177ff1 [Xiangrui Meng] address Matei's comments
be119fe [Xiangrui Meng] rename m/n to numRows/numCols for local matrix add tests for matrices
b881506 [Xiangrui Meng] rename SparkPCA/SVD to TallSkinnyPCA/SVD
e7d0d4a [Xiangrui Meng] move IndexedRDDMatrixRow to IndexedRowRDDMatrix
0d1491c [Xiangrui Meng] fix test errors
a85262a [Xiangrui Meng] rename RDDMatrixRow to IndexedRDDMatrixRow
b8b6ac3 [Xiangrui Meng] Remove old code
4cf679c [Xiangrui Meng] port pca to RowRDDMatrix, and add multiply and covariance
7836e2f [Xiangrui Meng] initial refactoring of matrices backed by RDDs
This is a patch to address @mateiz 's comment in https://github.com/apache/spark/pull/245
MLUtils#loadLibSVMData uses an anonymous function for the label parser. Java users won't like it. So I make a trait for LabelParser and provide two implementations: binary and multiclass.
Author: Xiangrui Meng <meng@databricks.com>
Closes#345 from mengxr/label-parser and squashes the following commits:
ac44409 [Xiangrui Meng] use singleton objects for label parsers
3b1a7c6 [Xiangrui Meng] add tests for label parsers
c2e571c [Xiangrui Meng] rename LabelParser.apply to LabelParser.parse use extends for singleton
11c94e0 [Xiangrui Meng] add return types
7f8eb36 [Xiangrui Meng] change labelParser from annoymous function to trait
Author: Holden Karau <holden@pigscanfly.ca>
Closes#242 from holdenk/spark-1320-cogroupandgroupshouldpassiterator and squashes the following commits:
f289536 [Holden Karau] Fix bad merge, should have been Iterable rather than Iterator
77048f8 [Holden Karau] Fix merge up to master
d3fe909 [Holden Karau] use toSeq instead
7a092a3 [Holden Karau] switch resultitr to resultiterable
eb06216 [Holden Karau] maybe I should have had a coffee first. use correct import for guava iterables
c5075aa [Holden Karau] If guava 14 had iterables
2d06e10 [Holden Karau] Fix Java 8 cogroup tests for the new API
11e730c [Holden Karau] Fix streaming tests
66b583d [Holden Karau] Fix the core test suite to compile
4ed579b [Holden Karau] Refactor from iterator to iterable
d052c07 [Holden Karau] Python tests now pass with iterator pandas
3bcd81d [Holden Karau] Revert "Try and make pickling list iterators work"
cd1e81c [Holden Karau] Try and make pickling list iterators work
c60233a [Holden Karau] Start investigating moving to iterators for python API like the Java/Scala one. tl;dr: We will have to write our own iterator since the default one doesn't pickle well
88a5cef [Holden Karau] Fix cogroup test in JavaAPISuite for streaming
a5ee714 [Holden Karau] oops, was checking wrong iterator
e687f21 [Holden Karau] Fix groupbykey test in JavaAPISuite of streaming
ec8cc3e [Holden Karau] Fix test issues\!
4b0eeb9 [Holden Karau] Switch cast in PairDStreamFunctions
fa395c9 [Holden Karau] Revert "Add a join based on the problem in SVD"
ec99e32 [Holden Karau] Revert "Revert this but for now put things in list pandas"
b692868 [Holden Karau] Revert
7e533f7 [Holden Karau] Fix the bug
8a5153a [Holden Karau] Revert me, but we have some stuff to debug
b4e86a9 [Holden Karau] Add a join based on the problem in SVD
c4510e2 [Holden Karau] Revert this but for now put things in list pandas
b4e0b1d [Holden Karau] Fix style issues
71e8b9f [Holden Karau] I really need to stop calling size on iterators, it is the path of sadness.
b1ae51a [Holden Karau] Fix some of the types in the streaming JavaAPI suite. Probably still needs more work
37888ec [Holden Karau] core/tests now pass
249abde [Holden Karau] org.apache.spark.rdd.PairRDDFunctionsSuite passes
6698186 [Holden Karau] Revert "I think this might be a bad rabbit hole. Started work to make CoGroupedRDD use iterator and then went crazy"
fe992fe [Holden Karau] hmmm try and fix up basic operation suite
172705c [Holden Karau] Fix Java API suite
caafa63 [Holden Karau] I think this might be a bad rabbit hole. Started work to make CoGroupedRDD use iterator and then went crazy
88b3329 [Holden Karau] Fix groupbykey to actually give back an iterator
4991af6 [Holden Karau] Fix some tests
be50246 [Holden Karau] Calling size on an iterator is not so good if we want to use it after
687ffbc [Holden Karau] This is the it compiles point of replacing Seq with Iterator and JList with JIterator in the groupby and cogroup signatures
In PR https://github.com/apache/spark/pull/117, we added dense/sparse vector data model and updated KMeans to support sparse input. This PR is to replace all other `Array[Double]` usage by `Vector` in generalized linear models (GLMs) and Naive Bayes. Major changes:
1. `LabeledPoint` becomes `LabeledPoint(Double, Vector)`.
2. Methods that accept `RDD[Array[Double]]` now accept `RDD[Vector]`. We cannot support both in an elegant way because of type erasure.
3. Mark 'createModel' and 'predictPoint' protected because they are not for end users.
4. Add libSVMFile to MLContext.
5. NaiveBayes can accept arbitrary labels (introducing a breaking change to Python's `NaiveBayesModel`).
6. Gradient computation no longer creates temp vectors.
7. Column normalization and centering are removed from Lasso and Ridge because the operation will densify the data. Simple feature transformation can be done before training.
TODO:
1. ~~Use axpy when possible.~~
2. ~~Optimize Naive Bayes.~~
Author: Xiangrui Meng <meng@databricks.com>
Closes#245 from mengxr/vector and squashes the following commits:
eb6e793 [Xiangrui Meng] move libSVMFile to MLUtils and rename to loadLibSVMData
c26c4fc [Xiangrui Meng] update DecisionTree to use RDD[Vector]
11999c7 [Xiangrui Meng] Merge branch 'master' into vector
f7da54b [Xiangrui Meng] add minSplits to libSVMFile
da25e24 [Xiangrui Meng] revert the change to default addIntercept because it might change the behavior of existing code without warning
493f26f [Xiangrui Meng] Merge branch 'master' into vector
7c1bc01 [Xiangrui Meng] add a TODO to NB
b9b7ef7 [Xiangrui Meng] change default value of addIntercept to false
b01df54 [Xiangrui Meng] allow to change or clear threshold in LR and SVM
4addc50 [Xiangrui Meng] merge master
4ca5b1b [Xiangrui Meng] remove normalization from Lasso and update tests
f04fe8a [Xiangrui Meng] remove normalization from RidgeRegression and update tests
d088552 [Xiangrui Meng] use static constructor for MLContext
6f59eed [Xiangrui Meng] update libSVMFile to determine number of features automatically
3432e84 [Xiangrui Meng] update NaiveBayes to support sparse data
0f8759b [Xiangrui Meng] minor updates to NB
b11659c [Xiangrui Meng] style update
78c4671 [Xiangrui Meng] add libSVMFile to MLContext
f0fe616 [Xiangrui Meng] add a test for sparse linear regression
44733e1 [Xiangrui Meng] use in-place gradient computation
e981396 [Xiangrui Meng] use axpy in Updater
db808a1 [Xiangrui Meng] update JavaLR example
befa592 [Xiangrui Meng] passed scala/java tests
75c83a4 [Xiangrui Meng] passed test compile
1859701 [Xiangrui Meng] passed compile
834ada2 [Xiangrui Meng] optimized MLUtils.computeStats update some ml algorithms to use Vector (cont.)
135ab72 [Xiangrui Meng] merge glm
0e57aa4 [Xiangrui Meng] update Lasso and RidgeRegression to parse the weights correctly from GLM mark createModel protected mark predictPoint protected
d7f629f [Xiangrui Meng] fix a bug in GLM when intercept is not used
3f346ba [Xiangrui Meng] update some ml algorithms to use Vector
Joint work with @hirakendu, @etrain, @atalwalkar and @harsha2010.
Key features:
+ Supports binary classification and regression
+ Supports gini, entropy and variance for information gain calculation
+ Supports both continuous and categorical features
The algorithm has gone through several development iterations over the last few months leading to a highly optimized implementation. Optimizations include:
1. Level-wise training to reduce passes over the entire dataset.
2. Bin-wise split calculation to reduce computation overhead.
3. Aggregation over partitions before combining to reduce communication overhead.
Author: Manish Amde <manish9ue@gmail.com>
Author: manishamde <manish9ue@gmail.com>
Author: Xiangrui Meng <meng@databricks.com>
Closes#79 from manishamde/tree and squashes the following commits:
1e8c704 [Manish Amde] remove numBins field in the Strategy class
7d54b4f [manishamde] Merge pull request #4 from mengxr/dtree
f536ae9 [Xiangrui Meng] another pass on code style
e1dd86f [Manish Amde] implementing code style suggestions
62dc723 [Manish Amde] updating javadoc and converting helper methods to package private to allow unit testing
201702f [Manish Amde] making some more methods private
f963ef5 [Manish Amde] making methods private
c487e6a [manishamde] Merge pull request #1 from mengxr/dtree
24500c5 [Xiangrui Meng] minor style updates
4576b64 [Manish Amde] documentation and for to while loop conversion
ff363a7 [Manish Amde] binary search for bins and while loop for categorical feature bins
632818f [Manish Amde] removing threshold for classification predict method
2116360 [Manish Amde] removing dummy bin calculation for categorical variables
6068356 [Manish Amde] ensuring num bins is always greater than max number of categories
62c2562 [Manish Amde] fixing comment indentation
ad1fc21 [Manish Amde] incorporated mengxr's code style suggestions
d1ef4f6 [Manish Amde] more documentation
794ff4d [Manish Amde] minor improvements to docs and style
eb8fcbe [Manish Amde] minor code style updates
cd2c2b4 [Manish Amde] fixing code style based on feedback
63e786b [Manish Amde] added multiple train methods for java compatability
d3023b3 [Manish Amde] adding more docs for nested methods
84f85d6 [Manish Amde] code documentation
9372779 [Manish Amde] code style: max line lenght <= 100
dd0c0d7 [Manish Amde] minor: some docs
0dd7659 [manishamde] basic doc
5841c28 [Manish Amde] unit tests for categorical features
f067d68 [Manish Amde] minor cleanup
c0e522b [Manish Amde] updated predict and split threshold logic
b09dc98 [Manish Amde] minor refactoring
6b7de78 [Manish Amde] minor refactoring and tests
d504eb1 [Manish Amde] more tests for categorical features
dbb7ac1 [Manish Amde] categorical feature support
6df35b9 [Manish Amde] regression predict logic
53108ed [Manish Amde] fixing index for highest bin
e23c2e5 [Manish Amde] added regression support
c8f6d60 [Manish Amde] adding enum for feature type
b0e3e76 [Manish Amde] adding enum for feature type
154aa77 [Manish Amde] enums for configurations
733d6dd [Manish Amde] fixed tests
02c595c [Manish Amde] added command line parsing
98ec8d5 [Manish Amde] tree building and prediction logic
b0eb866 [Manish Amde] added logic to handle leaf nodes
80e8c66 [Manish Amde] working version of multi-level split calculation
4798aae [Manish Amde] added gain stats class
dad0afc [Manish Amde] decison stump functionality working
03f534c [Manish Amde] some more tests
0012a77 [Manish Amde] basic stump working
8bca1e2 [Manish Amde] additional code for creating intermediate RDD
92cedce [Manish Amde] basic building blocks for intermediate RDD calculation. untested.
cd53eae [Manish Amde] skeletal framework
GLM needs to check addIntercept for intercept and weights. The current implementation always uses the first weight as intercept. Added a test for training without adding intercept.
JIRA: https://spark-project.atlassian.net/browse/SPARK-1327
Author: Xiangrui Meng <meng@databricks.com>
Closes#236 from mengxr/glm and squashes the following commits:
bcac1ac [Xiangrui Meng] add two tests to ensure {Lasso, Ridge}.setIntercept will throw an exceptions
a104072 [Xiangrui Meng] remove protected to be compatible with 0.9
0e57aa4 [Xiangrui Meng] update Lasso and RidgeRegression to parse the weights correctly from GLM mark createModel protected mark predictPoint protected
d7f629f [Xiangrui Meng] fix a bug in GLM when intercept is not used
Continue our discussions from https://github.com/apache/incubator-spark/pull/575
This PR is WIP because it depends on a SNAPSHOT version of breeze.
Per previous discussions and benchmarks, I switched to breeze for linear algebra operations. @dlwh and I made some improvements to breeze to keep its performance comparable to the bare-bone implementation, including norm computation and squared distance. This is why this PR needs to depend on a SNAPSHOT version of breeze.
@fommil , please find the notice of using netlib-core in `NOTICE`. This is following Apache's instructions on appropriate labeling.
I'm going to update this PR to include:
1. Fast distance computation: using `\|a\|_2^2 + \|b\|_2^2 - 2 a^T b` when it doesn't introduce too much numerical error. The squared norms are pre-computed. Otherwise, computing the distance between the center (dense) and a point (possibly sparse) always takes O(n) time.
2. Some numbers about the performance.
3. A released version of breeze. @dlwh, a minor release of breeze will help this PR get merged early. Do you mind sharing breeze's release plan? Thanks!
Author: Xiangrui Meng <meng@databricks.com>
Closes#117 from mengxr/sparse-kmeans and squashes the following commits:
67b368d [Xiangrui Meng] fix SparseVector.toArray
5eda0de [Xiangrui Meng] update NOTICE
67abe31 [Xiangrui Meng] move ArrayRDDs to mllib.rdd
1da1033 [Xiangrui Meng] remove dependency on commons-math3 and compute EPSILON directly
9bb1b31 [Xiangrui Meng] optimize SparseVector.toArray
226d2cd [Xiangrui Meng] update Java friendly methods in Vectors
238ba34 [Xiangrui Meng] add VectorRDDs with a converter from RDD[Array[Double]]
b28ba2f [Xiangrui Meng] add toArray to Vector
e69b10c [Xiangrui Meng] remove examples/JavaKMeans.java, which is replaced by mllib/examples/JavaKMeans.java
72bde33 [Xiangrui Meng] clean up code for distance computation
712cb88 [Xiangrui Meng] make Vectors.sparse Java friendly
27858e4 [Xiangrui Meng] update breeze version to 0.7
07c3cf2 [Xiangrui Meng] change Mahout to breeze in doc use a simple lower bound to avoid unnecessary distance computation
6f5cdde [Xiangrui Meng] fix a bug in filtering finished runs
42512f2 [Xiangrui Meng] Merge branch 'master' into sparse-kmeans
d6e6c07 [Xiangrui Meng] add predict(RDD[Vector]) to KMeansModel
42b4e50 [Xiangrui Meng] line feed at the end
a4ace73 [Xiangrui Meng] Merge branch 'fast-dist' into sparse-kmeans
3ed1a24 [Xiangrui Meng] add doc to BreezeVectorWithSquaredNorm
0107e19 [Xiangrui Meng] update NOTICE
87bc755 [Xiangrui Meng] tuned the KMeans code: changed some for loops to while, use view to avoid copying arrays
0ff8046 [Xiangrui Meng] update KMeans to use fastSquaredDistance
f355411 [Xiangrui Meng] add BreezeVectorWithSquaredNorm case class
ab74f67 [Xiangrui Meng] add fastSquaredDistance for KMeans
4e7d5ca [Xiangrui Meng] minor style update
07ffaf2 [Xiangrui Meng] add dense/sparse vector data models and conversions to/from breeze vectors use breeze to implement KMeans in order to support both dense and sparse data
# Principal Component Analysis
Computes the top k principal component coefficients for the m-by-n data matrix X. Rows of X correspond to observations and columns correspond to variables. The coefficient matrix is n-by-k. Each column of the coefficients return matrix contains coefficients for one principal component, and the columns are in descending order of component variance. This function centers the data and uses the singular value decomposition (SVD) algorithm.
## Testing
Tests included:
* All principal components
* Only top k principal components
* Dense SVD tests
* Dense/sparse matrix tests
The results are tested against MATLAB's pca: http://www.mathworks.com/help/stats/pca.html
## Documentation
Added to mllib-guide.md
## Example Usage
Added to examples directory under SparkPCA.scala
Author: Reza Zadeh <rizlar@gmail.com>
Closes#88 from rezazadeh/sparkpca and squashes the following commits:
e298700 [Reza Zadeh] reformat using IDE
3f23271 [Reza Zadeh] documentation and cleanup
b025ab2 [Reza Zadeh] documentation
e2667d4 [Reza Zadeh] assertMatrixApproximatelyEquals
3787bb4 [Reza Zadeh] stylin
c6ecc1f [Reza Zadeh] docs
aa2bbcb [Reza Zadeh] rename sparseToTallSkinnyDense
56975b0 [Reza Zadeh] docs
2df9bde [Reza Zadeh] docs update
8fb0015 [Reza Zadeh] rcond documentation
dbf7797 [Reza Zadeh] correct argument number
a9f1f62 [Reza Zadeh] documentation
4ce6caa [Reza Zadeh] style changes
9a56a02 [Reza Zadeh] use rcond relative to larget svalue
120f796 [Reza Zadeh] housekeeping
156ff78 [Reza Zadeh] string comprehension
2e1cf43 [Reza Zadeh] rename rcond
ea223a6 [Reza Zadeh] many style changes
f4002d7 [Reza Zadeh] more docs
bd53c7a [Reza Zadeh] proper accumulator
a8b5ecf [Reza Zadeh] Don't use for loops
0dc7980 [Reza Zadeh] filter zeros in sparse
6115610 [Reza Zadeh] More documentation
36d51e8 [Reza Zadeh] use JBLAS for UVS^-1 computation
bc4599f [Reza Zadeh] configurable rcond
86f7515 [Reza Zadeh] compute per parition, use while
09726b3 [Reza Zadeh] more style changes
4195e69 [Reza Zadeh] private, accumulator
17002be [Reza Zadeh] style changes
4ba7471 [Reza Zadeh] style change
f4982e6 [Reza Zadeh] Use dense matrix in example
2828d28 [Reza Zadeh] optimizations: normalize once, use inplace ops
72c9fa1 [Reza Zadeh] rename DenseMatrix to TallSkinnyDenseMatrix, lean
f807be9 [Reza Zadeh] fix typo
2d7ccde [Reza Zadeh] Array interface for dense svd and pca
cd290fa [Reza Zadeh] provide RDD[Array[Double]] support
398d123 [Reza Zadeh] style change
55abbfa [Reza Zadeh] docs fix
ef29644 [Reza Zadeh] bad chnage undo
472566e [Reza Zadeh] all files from old pr
555168f [Reza Zadeh] initial files
In implicit ALS computation, the user or product factor is used twice in each iteration. Caching can certainly help accelerate the computation. I saw the running time decreased by ~70% for implicit ALS on the movielens data.
I also made the following changes:
1. Change `YtYb` type from `Broadcast[Option[DoubleMatrix]]` to `Option[Broadcast[DoubleMatrix]]`, so we don't need to broadcast None in explicit computation.
2. Mark methods `computeYtY`, `unblockFactors`, `updateBlock`, and `updateFeatures private`. Users do not need those methods.
3. Materialize the final matrix factors before returning the model. It allows us to clean up other cached RDDs before returning the model. I do not have a better solution here, so I use `RDD.count()`.
JIRA: https://spark-project.atlassian.net/browse/SPARK-1266
Author: Xiangrui Meng <meng@databricks.com>
Closes#165 from mengxr/als and squashes the following commits:
c9676a6 [Xiangrui Meng] add a comment about the last products.persist
d3a88aa [Xiangrui Meng] change implicitPrefs match to if ... else ...
63862d6 [Xiangrui Meng] persist factors in implicit ALS
The current implementation uses `Array(1.0, features: _*)` to construct a new array with intercept. This is not efficient for big arrays because `Array.apply` uses a for loop that iterates over the arguments. `Array.+:` is a better choice here.
Also, I don't see a reason to set initial weights to ones. So I set them to zeros.
JIRA: https://spark-project.atlassian.net/browse/SPARK-1260
Author: Xiangrui Meng <meng@databricks.com>
Closes#161 from mengxr/sgd and squashes the following commits:
b5cfc53 [Xiangrui Meng] set default weights to zeros
a1439c2 [Xiangrui Meng] faster construction of features with intercept
Computing YtY can be implemented using BLAS's DSPR operations instead of generating y_i y_i^T and then combining them. The latter generates many k-by-k matrices. On the movielens data, this change improves the performance by 10-20%. The algorithm remains the same, verified by computing RMSE on the movielens data.
To compare the results, I also added an option to set a random seed in ALS.
JIRA:
1. https://spark-project.atlassian.net/browse/SPARK-1237
2. https://spark-project.atlassian.net/browse/SPARK-1238
Author: Xiangrui Meng <meng@databricks.com>
Closes#131 from mengxr/als and squashes the following commits:
ed00432 [Xiangrui Meng] minor changes
d984623 [Xiangrui Meng] minor changes
2fc1641 [Xiangrui Meng] remove commented code
4c7cde2 [Xiangrui Meng] allow specifying a random seed in ALS
200bef0 [Xiangrui Meng] optimize computeYtY and updateBlock
https://spark-project.atlassian.net/browse/SPARK-1160
reported by @mateiz: "It's redundant with collect() and the name doesn't make sense in Java, where we return a List (we can't return an array due to the way Java generics work). It's also missing in Python."
In this patch, I deprecated the method and changed the source files using it by replacing toArray with collect() directly
Author: CodingCat <zhunansjtu@gmail.com>
Closes#105 from CodingCat/SPARK-1060 and squashes the following commits:
286f163 [CodingCat] deprecate in JavaRDDLike
ee17b4e [CodingCat] add message and since
2ff7319 [CodingCat] deprecate toArray in RDD
Ported from https://github.com/apache/incubator-spark/pull/633
In runMiniBatchSGD, the regVal (for 1st iter) should be initialized
as sum of sqrt of weights if it's L2 update; for L1 update, the same logic is followed.
It maybe not be important here for SGD since the updater doesn't take the loss
as parameter to find the new weights. But it will give us the correct history of loss.
However, for LBFGS optimizer we implemented, the correct loss with regVal is crucial to
find the new weights.
Author: DB Tsai <dbtsai@alpinenow.com>
Closes#40 from dbtsai/dbtsai-smallRegValFix and squashes the following commits:
77d47da [DB Tsai] In runMiniBatchSGD, the regVal (for 1st iter) should be initialized as sum of sqrt of weights if it's L2 update; for L1 update, the same logic is followed.
There's a step in implicit ALS where the matrix `Yt * Y` is computed. It's computed as the sum of matrices; an f x f matrix is created for each of n user/item rows in a partition. In `ALS.scala:214`:
```
factors.flatMapValues{ case factorArray =>
factorArray.map{ vector =>
val x = new DoubleMatrix(vector)
x.mmul(x.transpose())
}
}.reduceByKeyLocally((a, b) => a.addi(b))
.values
.reduce((a, b) => a.addi(b))
```
Completely correct, but there's a subtle but quite large memory problem here. map() is going to create all of these matrices in memory at once, when they don't need to ever all exist at the same time.
For example, if a partition has n = 100000 rows, and f = 200, then this intermediate product requires 32GB of heap. The computation will never work unless you can cough up workers with (more than) that much heap.
Fortunately there's a trivial change that fixes it; just add `.view` in there.
Author: Sean Owen <sowen@cloudera.com>
Closes#629 from srowen/ALSMatrixAllocationOptimization and squashes the following commits:
062cda9 [Sean Owen] Update style per review comments
e9a5d63 [Sean Owen] Avoid unnecessary out of memory situation by not simultaneously allocating lots of matrices
I'm back with another less trivial suggestion for ALS:
In ALS for implicit feedback, input values are treated as weights on squared-errors in a loss function (or rather, the weight is a simple function of the input r, like c = 1 + alpha*r). The paper on which it's based assumes that the input is positive. Indeed, if the input is negative, it will create a negative weight on squared-errors, which causes things to go haywire. The optimization will try to make the error in a cell as large possible, and the result is silently bogus.
There is a good use case for negative input values though. Implicit feedback is usually collected from signals of positive interaction like a view or like or buy, but equally, can come from "not interested" signals. The natural representation is negative values.
The algorithm can be extended quite simply to provide a sound interpretation of these values: negative values should encourage the factorization to come up with 0 for cells with large negative input values, just as much as positive values encourage it to come up with 1.
The implications for the algorithm are simple:
* the confidence function value must not be negative, and so can become 1 + alpha*|r|
* the matrix P should have a value 1 where the input R is _positive_, not merely where it is non-zero. Actually, that's what the paper already says, it's just that we can't assume P = 1 when a cell in R is specified anymore, since it may be negative
This in turn entails just a few lines of code change in `ALS.scala`:
* `rs(i)` becomes `abs(rs(i))`
* When constructing `userXy(us(i))`, it's implicitly only adding where P is 1. That had been true for any us(i) that is iterated over, before, since these are exactly the ones for which P is 1. But now P is zero where rs(i) <= 0, and should not be added
I think it's a safe change because:
* It doesn't change any existing behavior (unless you're using negative values, in which case results are already borked)
* It's the simplest direct extension of the paper's algorithm
* (I've used it to good effect in production FWIW)
Tests included.
I tweaked minor things en route:
* `ALS.scala` javadoc writes "R = Xt*Y" when the paper and rest of code defines it as "R = X*Yt"
* RMSE in the ALS tests uses a confidence-weighted mean, but the denominator is not actually sum of weights
Excuse my Scala style; I'm sure it needs tweaks.
Author: Sean Owen <sowen@cloudera.com>
Closes#500 from srowen/ALSNegativeImplicitInput and squashes the following commits:
cf902a9 [Sean Owen] Support negative implicit input in ALS
953be1c [Sean Owen] Make weighted RMSE in ALS test actually weighted; adjust comment about R = X*Yt
url of "Collaborative Filtering for Implicit Feedback Datasets" is invalid now. A new url is provided. http://research.yahoo.com/files/HuKorenVolinsky-ICDM08.pdf
Author: Chen Chao <crazyjvm@gmail.com>
Closes#619 from CrazyJvm/master and squashes the following commits:
a0b54e4 [Chen Chao] change url to IEEE
9e0e9f0 [Chen Chao] correct spell mistale
fcfab5d [Chen Chao] wrap line to to fit within 100 chars
590d56e [Chen Chao] url error
new MLlib documentation for optimization, regression and classification
new documentation with tex formulas, hopefully improving usability and reproducibility of the offered MLlib methods.
also did some minor changes in the code for consistency. scala tests pass.
this is the rebased branch, i deleted the old PR
jira:
https://spark-project.atlassian.net/browse/MLLIB-19
Author: Martin Jaggi <m.jaggi@gmail.com>
Closes#566 and squashes the following commits:
5f0f31e [Martin Jaggi] line wrap at 100 chars
4e094fb [Martin Jaggi] better description of GradientDescent
1d6965d [Martin Jaggi] remove broken url
ea569c3 [Martin Jaggi] telling what updater actually does
964732b [Martin Jaggi] lambda R() in documentation
a6c6228 [Martin Jaggi] better comments in SGD code for regression
b32224a [Martin Jaggi] new optimization documentation
d5dfef7 [Martin Jaggi] new classification and regression documentation
b07ead6 [Martin Jaggi] correct scaling for MSE loss
ba6158c [Martin Jaggi] use d for the number of features
bab2ed2 [Martin Jaggi] renaming LeastSquaresGradient
Refactor RDD sampling and add randomSplit to RDD (update)
Replace SampledRDD by PartitionwiseSampledRDD, which accepts a RandomSampler instance as input. The current sample with/without replacement can be easily integrated via BernoulliSampler and PoissonSampler. The benefits are:
1) RDD.randomSplit is implemented in the same way, related to https://github.com/apache/incubator-spark/pull/513
2) Stratified sampling and importance sampling can be implemented in the same manner as well.
Unit tests are included for samplers and RDD.randomSplit.
This should performance better than my previous request where the BernoulliSampler creates many Iterator instances:
https://github.com/apache/incubator-spark/pull/513
Author: Xiangrui Meng <meng@databricks.com>
== Merge branch commits ==
commit e8ce957e5f0a600f2dec057924f4a2ca6adba373
Author: Xiangrui Meng <meng@databricks.com>
Date: Mon Feb 3 12:21:08 2014 -0800
more docs to PartitionwiseSampledRDD
commit fbb4586d0478ff638b24bce95f75ff06f713d43b
Author: Xiangrui Meng <meng@databricks.com>
Date: Mon Feb 3 00:44:23 2014 -0800
move XORShiftRandom to util.random and use it in BernoulliSampler
commit 987456b0ee8612fd4f73cb8c40967112dc3c4c2d
Author: Xiangrui Meng <meng@databricks.com>
Date: Sat Feb 1 11:06:59 2014 -0800
relax assertions in SortingSuite because the RangePartitioner has large variance in this case
commit 3690aae416b2dc9b2f9ba32efa465ba7948477f4
Author: Xiangrui Meng <meng@databricks.com>
Date: Sat Feb 1 09:56:28 2014 -0800
test split ratio of RDD.randomSplit
commit 8a410bc933a60c4d63852606f8bbc812e416d6ae
Author: Xiangrui Meng <meng@databricks.com>
Date: Sat Feb 1 09:25:22 2014 -0800
add a test to ensure seed distribution and minor style update
commit ce7e866f674c30ab48a9ceb09da846d5362ab4b6
Author: Xiangrui Meng <meng@databricks.com>
Date: Fri Jan 31 18:06:22 2014 -0800
minor style change
commit 750912b4d77596ed807d361347bd2b7e3b9b7a74
Author: Xiangrui Meng <meng@databricks.com>
Date: Fri Jan 31 18:04:54 2014 -0800
fix some long lines
commit c446a25c38d81db02821f7f194b0ce5ab4ed7ff5
Author: Xiangrui Meng <meng@databricks.com>
Date: Fri Jan 31 17:59:59 2014 -0800
add complement to BernoulliSampler and minor style changes
commit dbe2bc2bd888a7bdccb127ee6595840274499403
Author: Xiangrui Meng <meng@databricks.com>
Date: Fri Jan 31 17:45:08 2014 -0800
switch to partition-wise sampling for better performance
commit a1fca5232308feb369339eac67864c787455bb23
Merge: ac712e4 cf6128f
Author: Xiangrui Meng <meng@databricks.com>
Date: Fri Jan 31 16:33:09 2014 -0800
Merge branch 'sample' of github.com:mengxr/incubator-spark into sample
commit cf6128fb672e8c589615adbd3eaa3cbdb72bd461
Author: Xiangrui Meng <meng@databricks.com>
Date: Sun Jan 26 14:40:07 2014 -0800
set SampledRDD deprecated in 1.0
commit f430f847c3df91a3894687c513f23f823f77c255
Author: Xiangrui Meng <meng@databricks.com>
Date: Sun Jan 26 14:38:59 2014 -0800
update code style
commit a8b5e2021a9204e318c80a44d00c5c495f1befb6
Author: Xiangrui Meng <meng@databricks.com>
Date: Sun Jan 26 12:56:27 2014 -0800
move package random to util.random
commit ab0fa2c4965033737a9e3a9bf0a59cbb0df6a6f5
Author: Xiangrui Meng <meng@databricks.com>
Date: Sun Jan 26 12:50:35 2014 -0800
add Apache headers and update code style
commit 985609fe1a55655ad11966e05a93c18c138a403d
Author: Xiangrui Meng <meng@databricks.com>
Date: Sun Jan 26 11:49:25 2014 -0800
add new lines
commit b21bddf29850a2c006a868869b8f91960a029322
Author: Xiangrui Meng <meng@databricks.com>
Date: Sun Jan 26 11:46:35 2014 -0800
move samplers to random.IndependentRandomSampler and add tests
commit c02dacb4a941618e434cefc129c002915db08be6
Author: Xiangrui Meng <meng@databricks.com>
Date: Sat Jan 25 15:20:24 2014 -0800
add RandomSampler
commit 8ff7ba3c5cf1fc338c29ae8b5fa06c222640e89c
Author: Xiangrui Meng <meng@databricks.com>
Date: Fri Jan 24 13:23:22 2014 -0800
init impl of IndependentlySampledRDD
Choose initial user/item vectors uniformly on the unit sphere
...rather than within the unit square to possibly avoid bias in the initial state and improve convergence.
The current implementation picks the N vector elements uniformly at random from [0,1). This means they all point into one quadrant of the vector space. As N gets just a little large, the vector tend strongly to point into the "corner", towards (1,1,1...,1). The vectors are not unit vectors either.
I suggest choosing the elements as Gaussian ~ N(0,1) and normalizing. This gets you uniform random choices on the unit sphere which is more what's of interest here. It has worked a little better for me in the past.
This is pretty minor but wanted to warm up suggesting a few tweaks to ALS.
Please excuse my Scala, pretty new to it.
Author: Sean Owen <sowen@cloudera.com>
== Merge branch commits ==
commit 492b13a7469e5a4ed7591ee8e56d8bd7570dfab6
Author: Sean Owen <sowen@cloudera.com>
Date: Mon Jan 27 08:05:25 2014 +0000
Style: spaces around binary operators
commit ce2b5b5a4fefa0356875701f668f01f02ba4d87e
Author: Sean Owen <sowen@cloudera.com>
Date: Sun Jan 19 22:50:03 2014 +0000
Generate factors with all positive components, per discussion in https://github.com/apache/incubator-spark/pull/460
commit b6f7a8a61643a8209e8bc662e8e81f2d15c710c7
Author: Sean Owen <sowen@cloudera.com>
Date: Sat Jan 18 15:54:42 2014 +0000
Choose initial user/item vectors uniformly on the unit sphere rather than within the unit square to possibly avoid bias in the initial state and improve convergence
Sparse SVD
# Singular Value Decomposition
Given an *m x n* matrix *A*, compute matrices *U, S, V* such that
*A = U * S * V^T*
There is no restriction on m, but we require n^2 doubles to fit in memory.
Further, n should be less than m.
The decomposition is computed by first computing *A^TA = V S^2 V^T*,
computing svd locally on that (since n x n is small),
from which we recover S and V.
Then we compute U via easy matrix multiplication
as *U = A * V * S^-1*
Only singular vectors associated with the largest k singular values
If there are k such values, then the dimensions of the return will be:
* *S* is *k x k* and diagonal, holding the singular values on diagonal.
* *U* is *m x k* and satisfies U^T*U = eye(k).
* *V* is *n x k* and satisfies V^TV = eye(k).
All input and output is expected in sparse matrix format, 0-indexed
as tuples of the form ((i,j),value) all in RDDs.
# Testing
Tests included. They test:
- Decomposition promise (A = USV^T)
- For small matrices, output is compared to that of jblas
- Rank 1 matrix test included
- Full Rank matrix test included
- Middle-rank matrix forced via k included
# Example Usage
import org.apache.spark.SparkContext
import org.apache.spark.mllib.linalg.SVD
import org.apache.spark.mllib.linalg.SparseMatrix
import org.apache.spark.mllib.linalg.MatrixyEntry
// Load and parse the data file
val data = sc.textFile("mllib/data/als/test.data").map { line =>
val parts = line.split(',')
MatrixEntry(parts(0).toInt, parts(1).toInt, parts(2).toDouble)
}
val m = 4
val n = 4
// recover top 1 singular vector
val decomposed = SVD.sparseSVD(SparseMatrix(data, m, n), 1)
println("singular values = " + decomposed.S.data.toArray.mkString)
# Documentation
Added to docs/mllib-guide.md
Add Naive Bayes to Python MLlib, and some API fixes
- Added a Python wrapper for Naive Bayes
- Updated the Scala Naive Bayes to match the style of our other
algorithms better and in particular make it easier to call from Java
(added builder pattern, removed default value in train method)
- Updated Python MLlib functions to not require a SparkContext; we can
get that from the RDD the user gives
- Added a toString method in LabeledPoint
- Made the Python MLlib tests run as part of run-tests as well (before
they could only be run individually through each file)
- Added a Python wrapper for Naive Bayes
- Updated the Scala Naive Bayes to match the style of our other
algorithms better and in particular make it easier to call from Java
(added builder pattern, removed default value in train method)
- Updated Python MLlib functions to not require a SparkContext; we can
get that from the RDD the user gives
- Added a toString method in LabeledPoint
- Made the Python MLlib tests run as part of run-tests as well (before
they could only be run individually through each file)
standard Naive Bayes classifier
Has implemented the standard Naive Bayes classifier. This is an updated version of #288, which is closed because of misoperations.
* Arguments renamed according to Ameet's suggestion
* Using DoubleMatrix instead of Array[Double] in computation
* Removed arguments C (kinds of label) and D (dimension of feature vector) from NaiveBayes.train()
* Replaced reduceByKey with foldByKey to avoid modifying original input data
- Got rid of global SparkContext.globalConf
- Pass SparkConf to serializers and compression codecs
- Made SparkConf public instead of private[spark]
- Improved API of SparkContext and SparkConf
- Switched executor environment vars to be passed through SparkConf
- Fixed some places that were still using system properties
- Fixed some tests, though others are still failing
This still fails several tests in core, repl and streaming, likely due
to properties not being set or cleared correctly (some of the tests run
fine in isolation).
To run unit test, start SBT console and type:
compile
test-only org.apache.spark.util.XORShiftRandomSuite
To run benchmark, type:
project core
console
Once the Scala console starts, type:
org.apache.spark.util.XORShiftRandom.benchmark(100000000)
This change adds Java examples and unit tests for all GLM algorithms
to make sure the MLLib interface works from Java. Changes include
- Introduce LabeledPoint and avoid using Doubles in train arguments
- Rename train to run in class methods
- Make the optimizer a member variable of GLM to make sure the builder
pattern works
- Changes ALS to accept RDD[Rating] instead of (Int, Int, Double) making it
easier to call from Java
- Renames class methods from `train` to `run` to enable static methods to be
called from Java.
- Add unit tests which check if both static / class methods can be called.
- Also add examples which port the main() function in ALS, KMeans to the
examples project.
Couple of minor changes to existing code:
- Add a toJavaRDD method in RDD to convert scala RDD to java RDD easily
- Workaround a bug where using double[] from Java leads to class cast exception in
KMeans init
This change refactors the structure of GLMs to use mixins which maintain
a similar interface to other ML lib algorithms. This change also creates
an Optimizer trait which allows GLMs to be extended to use other optimization
techniques.
- Changed LogisticRegression regularization parameter to 0
- Removed println from SVM predict function
- Fixed "Lasso" -> "SVM" in SVMGenerator
- Added comment in Updater.scala to indicate L1 regularization leads to
soft thresholding proximal function
