### What changes were proposed in this pull request?
Code in the PR generates random parameters for hyperparameter tuning. A discussion with Sean Owen can be found on the dev mailing list here:
http://apache-spark-developers-list.1001551.n3.nabble.com/Hyperparameter-Optimization-via-Randomization-td30629.html
All code is entirely my own work and I license the work to the project under the project’s open source license.
### Why are the changes needed?
Randomization can be a more effective techinique than a grid search since min/max points can fall between the grid and never be found. Randomisation is not so restricted although the probability of finding minima/maxima is dependent on the number of attempts.
Alice Zheng has an accessible description on how this technique works at https://www.oreilly.com/library/view/evaluating-machine-learning/9781492048756/ch04.html
Although there are Python libraries with more sophisticated techniques, not every Spark developer is using Python.
### Does this PR introduce _any_ user-facing change?
A new class (`ParamRandomBuilder.scala`) and its tests have been created but there is no change to existing code. This class offers an alternative to `ParamGridBuilder` and can be dropped into the code wherever `ParamGridBuilder` appears. Indeed, it extends `ParamGridBuilder` and is completely compatible with its interface. It merely adds one method that provides a range over which a hyperparameter will be randomly defined.
### How was this patch tested?
Tests `ParamRandomBuilderSuite.scala` and `RandomRangesSuite.scala` were added.
`ParamRandomBuilderSuite` is the analogue of the already existing `ParamGridBuilderSuite` which tests the user-facing interface.
`RandomRangesSuite` uses ScalaCheck to test the random ranges over which hyperparameters are distributed.
Closes#31535 from PhillHenry/ParamRandomBuilder.
Authored-by: Phillip Henry <PhillHenry@gmail.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
Move the datetime rebase SQL configs from the `legacy` namespace by:
1. Renaming of the existing rebase configs like `spark.sql.legacy.parquet.datetimeRebaseModeInRead` -> `spark.sql.parquet.datetimeRebaseModeInRead`.
2. Add the legacy configs as alternatives
3. Deprecate the legacy rebase configs.
### Why are the changes needed?
The rebasing SQL configs like `spark.sql.legacy.parquet.datetimeRebaseModeInRead` can be used not only for migration from previous Spark versions but also to read/write datatime columns saved by other systems/frameworks/libs. So, the configs shouldn't be considered as legacy configs.
### Does this PR introduce _any_ user-facing change?
Should not. Users will see a warning if they still use one of the legacy configs.
### How was this patch tested?
1. Manually checking new configs:
```scala
scala> spark.conf.get("spark.sql.parquet.datetimeRebaseModeInRead")
res0: String = EXCEPTION
scala> spark.conf.set("spark.sql.legacy.parquet.datetimeRebaseModeInRead", "LEGACY")
21/02/17 14:57:10 WARN SQLConf: The SQL config 'spark.sql.legacy.parquet.datetimeRebaseModeInRead' has been deprecated in Spark v3.2 and may be removed in the future. Use 'spark.sql.parquet.datetimeRebaseModeInRead' instead.
scala> spark.conf.get("spark.sql.parquet.datetimeRebaseModeInRead")
res2: String = LEGACY
```
2. By running a datetime rebasing test suite:
```
$ build/sbt "test:testOnly *ParquetRebaseDatetimeV1Suite"
```
Closes#31576 from MaxGekk/rebase-confs-alternatives.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR proposes to use `_create_udf` where we need to create `UserDefinedFunction` to maintain codes easier.
### Why are the changes needed?
For the better readability of codes and maintenance.
### Does this PR introduce _any_ user-facing change?
No, refactoring.
### How was this patch tested?
Ran the existing unittests. CI in this PR should test it out too.
Closes#31537 from HyukjinKwon/SPARK-34408.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Creating a Pandas dataframe via Apache Arrow currently can use twice as much memory as the final result, because during the conversion, both Pandas and Arrow retain a copy of the data. Arrow has a "self-destruct" mode now (Arrow >= 0.16) to avoid this, by freeing each column after conversion. This PR integrates support for this in toPandas, handling a couple of edge cases:
self_destruct has no effect unless the memory is allocated appropriately, which is handled in the Arrow serializer here. Essentially, the issue is that self_destruct frees memory column-wise, but Arrow record batches are oriented row-wise:
```
Record batch 0: allocation 0: column 0 chunk 0, column 1 chunk 0, ...
Record batch 1: allocation 1: column 0 chunk 1, column 1 chunk 1, ...
```
In this scenario, Arrow will drop references to all of column 0's chunks, but no memory will actually be freed, as the chunks were just slices of an underlying allocation. The PR copies each column into its own allocation so that memory is instead arranged as so:
```
Record batch 0: allocation 0 column 0 chunk 0, allocation 1 column 1 chunk 0, ...
Record batch 1: allocation 2 column 0 chunk 1, allocation 3 column 1 chunk 1, ...
```
The optimization is disabled by default, and can be enabled with the Spark SQL conf "spark.sql.execution.arrow.pyspark.selfDestruct.enabled" set to "true". We can't always apply this optimization because it's more likely to generate a dataframe with immutable buffers, which Pandas doesn't always handle well, and because it is slower overall (since it only converts one column at a time instead of in parallel).
### Why are the changes needed?
This lets us load larger datasets - in particular, with N bytes of memory, before we could never load a dataset bigger than N/2 bytes; now the overhead is more like N/1.25 or so.
### Does this PR introduce _any_ user-facing change?
Yes - it adds a new SQL conf "spark.sql.execution.arrow.pyspark.selfDestruct.enabled"
### How was this patch tested?
See the [mailing list](http://apache-spark-developers-list.1001551.n3.nabble.com/DISCUSS-Reducing-memory-usage-of-toPandas-with-Arrow-quot-self-destruct-quot-option-td30149.html) - it was tested with Python memory_profiler. Unit tests added to check memory within certain bounds and correctness with the option enabled.
Closes#29818 from lidavidm/spark-32953.
Authored-by: David Li <li.davidm96@gmail.com>
Signed-off-by: Bryan Cutler <cutlerb@gmail.com>
### What changes were proposed in this pull request?
This follows up #31160 to update score function in the document.
### Why are the changes needed?
Currently we use `f_classif`, `ch2`, `f_regression`, which sound to me the sklearn's naming. It is good to have it but I think it is nice if we have formal score function name with sklearn's ones.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
No, only doc change.
Closes#31531 from viirya/SPARK-34080-minor.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
In the PR, I propose new options for the Parquet datasource:
1. `datetimeRebaseMode`
2. `int96RebaseMode`
Both options influence on loading ancient dates and timestamps column values from parquet files. The `datetimeRebaseMode` option impacts on loading values of the `DATE`, `TIMESTAMP_MICROS` and `TIMESTAMP_MILLIS` types, `int96RebaseMode` impacts on loading of `INT96` timestamps.
The options support the same values as the SQL configs `spark.sql.legacy.parquet.datetimeRebaseModeInRead` and `spark.sql.legacy.parquet.int96RebaseModeInRead` namely;
- `"LEGACY"`, when an option is set to this value, Spark rebases dates/timestamps from the legacy hybrid calendar (Julian + Gregorian) to the Proleptic Gregorian calendar.
- `"CORRECTED"`, dates/timestamps are read AS IS from parquet files.
- `"EXCEPTION"`, when it is set as an option value, Spark will fail the reading if it sees ancient dates/timestamps that are ambiguous between the two calendars.
### Why are the changes needed?
1. New options will allow to load parquet files from at least two sources in different rebasing modes in the same query. For instance:
```scala
val df1 = spark.read.option("datetimeRebaseMode", "legacy").parquet(folder1)
val df2 = spark.read.option("datetimeRebaseMode", "corrected").parquet(folder2)
df1.join(df2, ...)
```
Before the changes, it is impossible because the SQL config `spark.sql.legacy.parquet.datetimeRebaseModeInRead` influences on both reads.
2. Mixing of Dataset/DataFrame and RDD APIs should become possible. Since SQL configs are not propagated through RDDs, the following code fails on ancient timestamps:
```scala
spark.conf.set("spark.sql.legacy.parquet.datetimeRebaseModeInRead", "legacy")
spark.read.parquet(folder).distinct.rdd.collect()
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
By running the modified test suites:
```
$ build/sbt "sql/test:testOnly *ParquetRebaseDatetimeV1Suite"
$ build/sbt "sql/test:testOnly *ParquetRebaseDatetimeV2Suite"
```
Closes#31489 from MaxGekk/parquet-rebase-options.
Authored-by: Max Gekk <max.gekk@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
The current implement of some DDL not unify the output and not pass the output properly to physical command.
Such as: The `ShowTables` output attributes `namespace`, but `ShowTablesCommand` output attributes `database`.
As the query plan, this PR pass the output attributes from `ShowTables` to `ShowTablesCommand`, `ShowTableExtended ` to `ShowTablesCommand`.
Take `show tables` and `show table extended like 'tbl'` as example.
The output before this PR:
`show tables`
|database|tableName|isTemporary|
-- | -- | --
| default| tbl| false|
If catalog is v2 session catalog, the output before this PR:
|namespace|tableName|
-- | --
| default| tbl
`show table extended like 'tbl'`
|database|tableName|isTemporary| information|
-- | -- | -- | --
| default| tbl| false|Database: default...|
The output after this PR:
`show tables`
|namespace|tableName|isTemporary|
-- | -- | --
| default| tbl| false|
`show table extended like 'tbl'`
|namespace|tableName|isTemporary| information|
-- | -- | -- | --
| default| tbl| false|Database: default...|
### Why are the changes needed?
This PR have benefits as follows:
First, Unify schema for the output of SHOW TABLES.
Second, pass the output attributes could keep the expr ID unchanged, so that avoid bugs when we apply more operators above the command output dataframe.
### Does this PR introduce _any_ user-facing change?
Yes.
The output schema of `SHOW TABLES` replace `database` by `namespace`.
### How was this patch tested?
Jenkins test.
Closes#31245 from beliefer/SPARK-34157.
Lead-authored-by: gengjiaan <gengjiaan@360.cn>
Co-authored-by: beliefer <beliefer@163.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
This changeset is published into the public domain.
### What changes were proposed in this pull request?
Some typos and syntax issues in docstrings and the output of `dev/lint-python` have been fixed.
### Why are the changes needed?
In some places, the documentation did not refer to parameters or classes by the full and correct name, potentially causing uncertainty in the reader or rendering issues in Sphinx. Also, a typo in the standard output of `dev/lint-python` was fixed.
### Does this PR introduce _any_ user-facing change?
Slight improvements in documentation, and in standard output of `dev/lint-python`.
### How was this patch tested?
Manual testing and `dev/lint-python` run. No new Sphinx warnings arise due to this change.
Closes#31401 from DavidToneian/SPARK-34300.
Authored-by: David Toneian <david@toneian.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR completes snake_case rule at functions APIs across the languages, see also SPARK-10621.
In more details, this PR:
- Adds `count_distinct` in Scala Python, and R, and document that `count_distinct` is encouraged. This was not deprecated because `countDistinct` is pretty commonly used. We could deprecate in the future releases.
- (Scala-specific) adds `typedlit` but doesn't deprecate `typedLit` which is arguably commonly used. Likewise, we could deprecate in the future releases.
- Deprecates and renames:
- `sumDistinct` -> `sum_distinct`
- `bitwiseNOT` -> `bitwise_not`
- `shiftLeft` -> `shiftleft` (matched with SQL name in `FunctionRegistry`)
- `shiftRight` -> `shiftright` (matched with SQL name in `FunctionRegistry`)
- `shiftRightUnsigned` -> `shiftrightunsigned` (matched with SQL name in `FunctionRegistry`)
- (Scala-specific) `callUDF` -> `call_udf`
### Why are the changes needed?
To keep the consistent naming in APIs.
### Does this PR introduce _any_ user-facing change?
Yes, it deprecates some APIs and add new renamed APIs as described above.
### How was this patch tested?
Unittests were added.
Closes#31408 from HyukjinKwon/SPARK-34306.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
1, use Guavaording instead of BoundedPriorityQueue;
2, use local variables;
3, avoid conversion: ml.vector -> mllib.vector
### Why are the changes needed?
this pr is about 30% faster than existing impl
### Does this PR introduce _any_ user-facing change?
NO
### How was this patch tested?
existing testsuites
Closes#31276 from zhengruifeng/w2v_findSynonyms_opt.
Authored-by: Ruifeng Zheng <ruifengz@foxmail.com>
Signed-off-by: Ruifeng Zheng <ruifengz@foxmail.com>
### What changes were proposed in this pull request?
Adds `NullType` support for Arrow executions.
### Why are the changes needed?
As Arrow supports null type, we can convert `NullType` between PySpark and pandas with Arrow enabled.
### Does this PR introduce _any_ user-facing change?
Yes, if a user has a DataFrame including `NullType`, it will be able to convert with Arrow enabled.
### How was this patch tested?
Added tests.
Closes#31285 from ueshin/issues/SPARK-33489/arrow_nulltype.
Authored-by: Takuya UESHIN <ueshin@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Added typing for keyword-only single argument udf overload.
### Why are the changes needed?
The intended use case is:
```
udf(returnType="string")
def f(x): ...
```
### Does this PR introduce _any_ user-facing change?
Yes - a new typing for udf is considered valid.
### How was this patch tested?
Existing tests.
Closes#31282 from pgrz/patch-1.
Authored-by: pgrz <grzegorski.piotr@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes to consider the case when [`inspect.currentframe()`](https://docs.python.org/3/library/inspect.html#inspect.currentframe) returns `None` because the underlyining Python implementation does not support frame.
### Why are the changes needed?
To be safer and potentially for the official support of other Python implementations in the future.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Manually tested via:
When frame is available:
```
vi tmp.py
```
```python
from inspect import *
lineno = getframeinfo(currentframe()).lineno + 1 if currentframe() is not None else 0
print(warnings.formatwarning(
"Failed to set memory limit: {0}".format(Exception("argh!")),
ResourceWarning,
__file__,
lineno),
file=sys.stderr)
```
```
python tmp.py
```
```
/.../tmp.py:3: ResourceWarning: Failed to set memory limit: argh!
print(warnings.formatwarning(
```
When frame is not available:
```
vi tmp.py
```
```python
from inspect import *
lineno = getframeinfo(currentframe()).lineno + 1 if None is not None else 0
print(warnings.formatwarning(
"Failed to set memory limit: {0}".format(Exception("argh!")),
ResourceWarning,
__file__,
lineno),
file=sys.stderr)
```
```
python tmp.py
```
```
/.../tmp.py:0: ResourceWarning: Failed to set memory limit: argh!
```
Closes#31239 from HyukjinKwon/SPARK-33730-followup.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
update the ParamValidators of `maxDepth`
### Why are the changes needed?
current impl of tree models only support maxDepth<=30
### Does this PR introduce _any_ user-facing change?
If `maxDepth`>30, fail quickly
### How was this patch tested?
existing testsuites
Closes#31163 from zhengruifeng/param_maxDepth_upbound.
Authored-by: Ruifeng Zheng <ruifengz@foxmail.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
This PR:
- Adds as small hierarchy of warnings to be used in PySpark applications. These extend built-in classes and top level `PySparkWarning`.
- Replaces `DeprecationWarnings` (intended for developers) with PySpark specific subclasses of `FutureWarning` (intended for end users).
### Why are the changes needed?
- To be more precise and add users additional control (in addition to standard module level filters) over PySpark warnings handling.
- Correct semantics (at the moment we use `DeprecationWarning` in user-facing API, but it is intended "for warnings about deprecated features when those warnings are intended for other Python developers").
### Does this PR introduce _any_ user-facing change?
Yes. Code can raise different type of warning than before.
### How was this patch tested?
Existing tests.
Closes#30985 from zero323/SPARK-33730.
Authored-by: zero323 <mszymkiewicz@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Add UnivariateFeatureSelector
### Why are the changes needed?
Have one UnivariateFeatureSelector, so we don't need to have three Feature Selectors.
### Does this PR introduce _any_ user-facing change?
Yes
```
selector = UnivariateFeatureSelector(featureCols=["x", "y", "z"], labelCol=["target"], featureType="categorical", labelType="continuous", selectorType="numTopFeatures", numTopFeatures=100)
```
Or
numTopFeatures
```
selector = UnivariateFeatureSelector(featureCols=["x", "y", "z"], labelCol=["target"], scoreFunction="f_classif", selectorType="numTopFeatures", numTopFeatures=100)
```
### How was this patch tested?
Add Unit test
Closes#31160 from huaxingao/UnivariateSelector.
Authored-by: Huaxin Gao <huaxing@us.ibm.com>
Signed-off-by: Weichen Xu <weichen.xu@databricks.com>
### What changes were proposed in this pull request?
This PR aims to strip auto-generated cast. The main logic is:
1. Add tag if Cast is specified by user.
2. Wrap `PrettyAttribute` in usePrettyExpression.
### Why are the changes needed?
Make sql consistent with dsl. Here is an inconsistent example before this PR:
```
-- output field name: FLOOR(1)
spark.emptyDataFrame.select(floor(lit(1)))
-- output field name: FLOOR(CAST(1 AS DOUBLE))
spark.sql("select floor(1)")
```
Note that, we don't remove the `Cast` so the auto-generated `Cast` can still work. The only changed place is `usePrettyExpression`, we use `PrettyAttribute` replace `Cast` to give a better sql string.
### Does this PR introduce _any_ user-facing change?
Yes, the default field name may change.
### How was this patch tested?
Add test and pass exists test.
Closes#31034 from ulysses-you/SPARK-33989.
Authored-by: ulysses-you <ulyssesyou18@gmail.com>
Signed-off-by: Wenchen Fan <wenchen@databricks.com>
### What changes were proposed in this pull request?
This PR is a follow-up of #29138 and #29195 to add more tests for `slice` function.
### Why are the changes needed?
The original PRs are missing tests with column-based arguments instead of literals.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Added tests and existing tests.
Closes#31159 from ueshin/issues/SPARK-32338/slice_tests.
Authored-by: Takuya UESHIN <ueshin@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR is a followup of https://github.com/apache/spark/pull/27406. It fixes the naming to match with Scala side.
Note that there are a bit of inconsistency already e.g.) `col`, `e`, `expr` and `column`. This part I did not change but other names like `zero` vs `initialValue` or `col1`/`col2` vs `left`/`right` looks unnecessary.
### Why are the changes needed?
To make the usage similar with Scala side, and for consistency.
### Does this PR introduce _any_ user-facing change?
No, this is not released yet.
### How was this patch tested?
GitHub Actions and Jenkins build will test it out.
Closes#31062 from HyukjinKwon/SPARK-30681.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes to upgrade cloudpickle from 1.5.0 to 1.6.0.
It virtually contains one fix:
4510be850d
From a cursory look, this isn't a regression, and not even properly supported in Python:
```python
>>> import pickle
>>> pickle.dumps({}.keys())
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: cannot pickle 'dict_keys' object
```
So it seems fine not to backport.
### Why are the changes needed?
To leverage bug fixes from the cloudpickle upstream.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Jenkins build and GitHub actions build will test it out.
Closes#31007 from HyukjinKwon/cloudpickle-upgrade.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
Follow up work for #30521, document the following behaviors in the API doc:
- Figure out the effects when configurations are (provider/partitionBy) conflicting with the existing table.
- Document the lack of functionality on creating a v2 table, and guide that the users should ensure a table is created in prior to avoid the behavior unintended/insufficient table is being created.
### Why are the changes needed?
We didn't have full support for the V2 table created in the API now. (TODO SPARK-33638)
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Document only.
Closes#30885 from xuanyuanking/SPARK-33659.
Authored-by: Yuanjian Li <yuanjian.li@databricks.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Reopened from https://github.com/apache/spark/pull/27525.
The exception messages for dstream.py when using windows were improved to be specific about what sliding duration is important.
### Why are the changes needed?
The batch interval of dstreams are improperly named as sliding windows. The term sliding window is also used to reference the new window of a dstream collected over a window of rdds in a parent dstream. We should probably fix the naming convention of sliding window used in the dstream class, but for now more this more explicit exception message may reduce confusion.
### Does this PR introduce any user-facing change?
No
### How was this patch tested?
It wasn't since this is only a change of the exception message
Closes#30871 from kykrueger/kykrueger-patch-1.
Authored-by: Kyle Krueger <kyle.s.krueger@gmail.com>
Signed-off-by: Dongjoon Hyun <dhyun@apple.com>
### What changes were proposed in this pull request?
This PR proposes to:
- Make doctests simpler to show the usage (since we're not running them now).
- Use the test utils to drop the tables if exists.
### Why are the changes needed?
Better docs and code readability.
### Does this PR introduce _any_ user-facing change?
No, dev-only. It includes some doc changes in unreleased branches.
### How was this patch tested?
Manually tested.
```bash
cd python
./run-tests --python-executable=python3.9,python3.8 --testnames "pyspark.sql.tests.test_streaming StreamingTests"
```
Closes#30873 from HyukjinKwon/SPARK-33836.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
### What changes were proposed in this pull request?
This PR proposes to expose `DataStreamReader.table` (SPARK-32885) and `DataStreamWriter.toTable` (SPARK-32896) to PySpark, which are the only way to read and write with table in Structured Streaming.
### Why are the changes needed?
Please refer SPARK-32885 and SPARK-32896 for rationalizations of these public APIs. This PR only exposes them to PySpark.
### Does this PR introduce _any_ user-facing change?
Yes, PySpark users will be able to read and write with table in Structured Streaming query.
### How was this patch tested?
Manually tested.
> v1 table
>> create table A and ingest to the table A
```
spark.sql("""
create table table_pyspark_parquet (
value long,
`timestamp` timestamp
) USING parquet
""")
df = spark.readStream.format('rate').option('rowsPerSecond', 100).load()
query = df.writeStream.toTable('table_pyspark_parquet', checkpointLocation='/tmp/checkpoint5')
query.lastProgress
query.stop()
```
>> read table A and ingest to the table B which doesn't exist
```
df2 = spark.readStream.table('table_pyspark_parquet')
query2 = df2.writeStream.toTable('table_pyspark_parquet_nonexist', format='parquet', checkpointLocation='/tmp/checkpoint2')
query2.lastProgress
query2.stop()
```
>> select tables
```
spark.sql("DESCRIBE TABLE table_pyspark_parquet").show()
spark.sql("SELECT * FROM table_pyspark_parquet").show()
spark.sql("DESCRIBE TABLE table_pyspark_parquet_nonexist").show()
spark.sql("SELECT * FROM table_pyspark_parquet_nonexist").show()
```
> v2 table (leveraging Apache Iceberg as it provides V2 table and custom catalog as well)
>> create table A and ingest to the table A
```
spark.sql("""
create table iceberg_catalog.default.table_pyspark_v2table (
value long,
`timestamp` timestamp
) USING iceberg
""")
df = spark.readStream.format('rate').option('rowsPerSecond', 100).load()
query = df.select('value', 'timestamp').writeStream.toTable('iceberg_catalog.default.table_pyspark_v2table', checkpointLocation='/tmp/checkpoint_v2table_1')
query.lastProgress
query.stop()
```
>> ingest to the non-exist table B
```
df2 = spark.readStream.format('rate').option('rowsPerSecond', 100).load()
query2 = df2.select('value', 'timestamp').writeStream.toTable('iceberg_catalog.default.table_pyspark_v2table_nonexist', checkpointLocation='/tmp/checkpoint_v2table_2')
query2.lastProgress
query2.stop()
```
>> ingest to the non-exist table C partitioned by `value % 10`
```
df3 = spark.readStream.format('rate').option('rowsPerSecond', 100).load()
df3a = df3.selectExpr('value', 'timestamp', 'value % 10 AS partition').repartition('partition')
query3 = df3a.writeStream.partitionBy('partition').toTable('iceberg_catalog.default.table_pyspark_v2table_nonexist_partitioned', checkpointLocation='/tmp/checkpoint_v2table_3')
query3.lastProgress
query3.stop()
```
>> select tables
```
spark.sql("DESCRIBE TABLE iceberg_catalog.default.table_pyspark_v2table").show()
spark.sql("SELECT * FROM iceberg_catalog.default.table_pyspark_v2table").show()
spark.sql("DESCRIBE TABLE iceberg_catalog.default.table_pyspark_v2table_nonexist").show()
spark.sql("SELECT * FROM iceberg_catalog.default.table_pyspark_v2table_nonexist").show()
spark.sql("DESCRIBE TABLE iceberg_catalog.default.table_pyspark_v2table_nonexist_partitioned").show()
spark.sql("SELECT * FROM iceberg_catalog.default.table_pyspark_v2table_nonexist_partitioned").show()
```
Closes#30835 from HeartSaVioR/SPARK-33836.
Lead-authored-by: Jungtaek Lim <kabhwan.opensource@gmail.com>
Co-authored-by: Jungtaek Lim (HeartSaVioR) <kabhwan.opensource@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR proposes to change python to python3 in several places missed.
### Why are the changes needed?
To use Python 3 by default safely.
### Does this PR introduce _any_ user-facing change?
Yes, it will uses `python3` as its default Python interpreter.
### How was this patch tested?
It was tested together in https://github.com/apache/spark/pull/30735. The test cases there will verify this change together.
Closes#30750 from HyukjinKwon/SPARK-32447.
Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR aims to update `master` branch version to 3.2.0-SNAPSHOT.
### Why are the changes needed?
Start to prepare Apache Spark 3.2.0.
### Does this PR introduce _any_ user-facing change?
N/A.
### How was this patch tested?
Pass the CIs.
Closes#30606 from dongjoon-hyun/SPARK-3.2.
Authored-by: Dongjoon Hyun <dongjoon@apache.org>
Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
### What changes were proposed in this pull request?
make CrossValidator/TrainValidateSplit/OneVsRest Reader/Writer support Python backend estimator/model
### Why are the changes needed?
Currently, pyspark support third-party library to define python backend estimator/evaluator, i.e., estimator that inherit `Estimator` instead of `JavaEstimator`, and only can be used in pyspark.
CrossValidator and TrainValidateSplit support tuning these python backend estimator,
but cannot support saving/load, becase CrossValidator and TrainValidateSplit writer implementation is use JavaMLWriter, which require to convert nested estimator and evaluator into java instance.
OneVsRest saving/load now only support java backend classifier due to similar issue.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Unit test.
Closes#30471 from WeichenXu123/support_pyio_tuning.
Authored-by: Weichen Xu <weichen.xu@databricks.com>
Signed-off-by: Weichen Xu <weichen.xu@databricks.com>
### What changes were proposed in this pull request?
`spark.buffer.size` not applied in driver from pyspark. In this PR I've fixed this issue.
### Why are the changes needed?
Apply the mentioned config on driver side.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Existing unit tests + manually.
Added the following code temporarily:
```
def local_connect_and_auth(port, auth_secret):
...
sock.connect(sa)
print("SPARK_BUFFER_SIZE: %d" % int(os.environ.get("SPARK_BUFFER_SIZE", 65536))) <- This is the addition
sockfile = sock.makefile("rwb", int(os.environ.get("SPARK_BUFFER_SIZE", 65536)))
...
```
Test:
```
#Compile Spark
echo "spark.buffer.size 10000" >> conf/spark-defaults.conf
$ ./bin/pyspark
Python 3.8.5 (default, Jul 21 2020, 10:48:26)
[Clang 11.0.3 (clang-1103.0.32.62)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
20/12/03 13:38:13 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
20/12/03 13:38:14 WARN SparkEnv: I/O encryption enabled without RPC encryption: keys will be visible on the wire.
Welcome to
____ __
/ __/__ ___ _____/ /__
_\ \/ _ \/ _ `/ __/ '_/
/__ / .__/\_,_/_/ /_/\_\ version 3.1.0-SNAPSHOT
/_/
Using Python version 3.8.5 (default, Jul 21 2020 10:48:26)
Spark context Web UI available at http://192.168.0.189:4040
Spark context available as 'sc' (master = local[*], app id = local-1606999094506).
SparkSession available as 'spark'.
>>> sc.setLogLevel("TRACE")
>>> sc.parallelize([0, 2, 3, 4, 6], 5).glom().collect()
...
SPARK_BUFFER_SIZE: 10000
...
[[0], [2], [3], [4], [6]]
>>>
```
Closes#30592 from gaborgsomogyi/SPARK-33629.
Authored-by: Gabor Somogyi <gabor.g.somogyi@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This is to update `labelsArray`'s since tag.
### Why are the changes needed?
The original change was backported to branch-3.0 for 3.0.2 version. So it is better to update the since tag to reflect the fact.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
N/A. Just tag change.
Closes#30582 from viirya/SPARK-33636-followup.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This is a followup to add missing `labelsArray` to PySpark `StringIndexer`.
### Why are the changes needed?
`labelsArray` is for multi-column case for `StringIndexer`. We should provide this accessor at PySpark side too.
### Does this PR introduce _any_ user-facing change?
Yes, `labelsArray` was missing in PySpark `StringIndexer` in Spark 3.0.
### How was this patch tested?
Unit test.
Closes#30579 from viirya/SPARK-22798-followup.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Fix: Pyspark ML Validator params in estimatorParamMaps may be lost after saving and reloading
When saving validator estimatorParamMaps, will check all nested stages in tuned estimator to get correct param parent.
Two typical cases to manually test:
~~~python
tokenizer = Tokenizer(inputCol="text", outputCol="words")
hashingTF = HashingTF(inputCol=tokenizer.getOutputCol(), outputCol="features")
lr = LogisticRegression()
pipeline = Pipeline(stages=[tokenizer, hashingTF, lr])
paramGrid = ParamGridBuilder() \
.addGrid(hashingTF.numFeatures, [10, 100]) \
.addGrid(lr.maxIter, [100, 200]) \
.build()
tvs = TrainValidationSplit(estimator=pipeline,
estimatorParamMaps=paramGrid,
evaluator=MulticlassClassificationEvaluator())
tvs.save(tvsPath)
loadedTvs = TrainValidationSplit.load(tvsPath)
# check `loadedTvs.getEstimatorParamMaps()` restored correctly.
~~~
~~~python
lr = LogisticRegression()
ova = OneVsRest(classifier=lr)
grid = ParamGridBuilder().addGrid(lr.maxIter, [100, 200]).build()
evaluator = MulticlassClassificationEvaluator()
tvs = TrainValidationSplit(estimator=ova, estimatorParamMaps=grid, evaluator=evaluator)
tvs.save(tvsPath)
loadedTvs = TrainValidationSplit.load(tvsPath)
# check `loadedTvs.getEstimatorParamMaps()` restored correctly.
~~~
### Why are the changes needed?
Bug fix.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Unit test.
Closes#30539 from WeichenXu123/fix_tuning_param_maps_io.
Authored-by: Weichen Xu <weichen.xu@databricks.com>
Signed-off-by: Ruifeng Zheng <ruifengz@foxmail.com>
### What changes were proposed in this pull request?
This replaces deprecated API usage in PySpark tests with the preferred APIs. These have been deprecated for some time and usage is not consistent within tests.
- https://docs.python.org/3/library/unittest.html#deprecated-aliases
### Why are the changes needed?
For consistency and eventual removal of deprecated APIs.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Existing tests
Closes#30557 from BryanCutler/replace-deprecated-apis-in-tests.
Authored-by: Bryan Cutler <cutlerb@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Add array_to_vector function for dataframe column
### Why are the changes needed?
Utility function for array to vector conversion.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
scala unit test & doctest.
Closes#30498 from WeichenXu123/array_to_vec.
Lead-authored-by: Weichen Xu <weichen.xu@databricks.com>
Co-authored-by: Hyukjin Kwon <gurwls223@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR intends to fix typos in the sub-modules:
* `R`
* `common`
* `dev`
* `mlib`
* `external`
* `project`
* `streaming`
* `resource-managers`
* `python`
Split per srowen https://github.com/apache/spark/pull/30323#issuecomment-728981618
NOTE: The misspellings have been reported at 706a726f87 (commitcomment-44064356)
### Why are the changes needed?
Misspelled words make it harder to read / understand content.
### Does this PR introduce _any_ user-facing change?
There are various fixes to documentation, etc...
### How was this patch tested?
No testing was performed
Closes#30402 from jsoref/spelling-R_common_dev_mlib_external_project_streaming_resource-managers_python.
Authored-by: Josh Soref <jsoref@users.noreply.github.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
There are some differences between Spark CSV, opencsv and commons-csv, the typical case are described in SPARK-33566, When there are both unescaped quotes and unescaped qualifier in value, the results of parsing are different.
The reason for the difference is Spark use `STOP_AT_DELIMITER` as default `UnescapedQuoteHandling` to build `CsvParser` and it not configurable.
On the other hand, opencsv and commons-csv use the parsing mechanism similar to `STOP_AT_CLOSING_QUOTE ` by default.
So this pr make `unescapedQuoteHandling` option configurable to get the same parsing result as opencsv and commons-csv.
### Why are the changes needed?
Make unescapedQuoteHandling option configurable when read CSV to make parsing more flexible。
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- Pass the Jenkins or GitHub Action
- Add a new case similar to that described in SPARK-33566
Closes#30518 from LuciferYang/SPARK-33566.
Authored-by: yangjie01 <yangjie01@baidu.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This PR adds the following functions (introduced in Scala API with SPARK-33061):
- `acosh`
- `asinh`
- `atanh`
to Python and R.
### Why are the changes needed?
Feature parity.
### Does this PR introduce _any_ user-facing change?
New functions.
### How was this patch tested?
New unit tests.
Closes#30501 from zero323/SPARK-33563.
Authored-by: zero323 <mszymkiewicz@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Spark creates local server to serialize several type of data for python. The python code tries to connect to the server, immediately after it's created but there are several system calls in between (this may change in each Spark version):
* getaddrinfo
* socket
* settimeout
* connect
Under some circumstances in heavy user environments these calls can be super slow (more than 15 seconds). These issues must be analyzed one-by-one but since these are system calls the underlying OS and/or DNS servers must be debugged and fixed. This is not trivial task and at the same time data processing must work somehow. In this PR I'm only intended to add a configuration possibility to increase the mentioned timeouts in order to be able to provide temporary workaround. The rootcause analysis is ongoing but I think this can vary in each case.
Because the server part doesn't contain huge amount of log entries to with one can measure time, I've added some.
### Why are the changes needed?
Provide workaround when localhost python server connection timeout appears.
### Does this PR introduce _any_ user-facing change?
Yes, new configuration added.
### How was this patch tested?
Existing unit tests + manual test.
```
#Compile Spark
echo "spark.io.encryption.enabled true" >> conf/spark-defaults.conf
echo "spark.python.authenticate.socketTimeout 10" >> conf/spark-defaults.conf
$ ./bin/pyspark
Python 3.8.5 (default, Jul 21 2020, 10:48:26)
[Clang 11.0.3 (clang-1103.0.32.62)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
20/11/20 10:17:03 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
20/11/20 10:17:03 WARN SparkEnv: I/O encryption enabled without RPC encryption: keys will be visible on the wire.
Welcome to
____ __
/ __/__ ___ _____/ /__
_\ \/ _ \/ _ `/ __/ '_/
/__ / .__/\_,_/_/ /_/\_\ version 3.1.0-SNAPSHOT
/_/
Using Python version 3.8.5 (default, Jul 21 2020 10:48:26)
Spark context Web UI available at http://192.168.0.189:4040
Spark context available as 'sc' (master = local[*], app id = local-1605863824276).
SparkSession available as 'spark'.
>>> sc.setLogLevel("TRACE")
>>> sc.parallelize([0, 2, 3, 4, 6], 5).glom().collect()
20/11/20 10:17:09 TRACE PythonParallelizeServer: Creating listening socket
20/11/20 10:17:09 TRACE PythonParallelizeServer: Setting timeout to 10 sec
20/11/20 10:17:09 TRACE PythonParallelizeServer: Waiting for connection on port 59726
20/11/20 10:17:09 TRACE PythonParallelizeServer: Connection accepted from address /127.0.0.1:59727
20/11/20 10:17:09 TRACE PythonParallelizeServer: Client authenticated
20/11/20 10:17:09 TRACE PythonParallelizeServer: Closing server
...
20/11/20 10:17:10 TRACE SocketFuncServer: Creating listening socket
20/11/20 10:17:10 TRACE SocketFuncServer: Setting timeout to 10 sec
20/11/20 10:17:10 TRACE SocketFuncServer: Waiting for connection on port 59735
20/11/20 10:17:10 TRACE SocketFuncServer: Connection accepted from address /127.0.0.1:59736
20/11/20 10:17:10 TRACE SocketFuncServer: Client authenticated
20/11/20 10:17:10 TRACE SocketFuncServer: Closing server
[[0], [2], [3], [4], [6]]
>>>
```
Closes#30389 from gaborgsomogyi/SPARK-33143.
Lead-authored-by: Gabor Somogyi <gabor.g.somogyi@gmail.com>
Co-authored-by: Hyukjin Kwon <gurwls223@gmail.com>
Co-authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
Two new options, _modifiiedBefore_ and _modifiedAfter_, is provided expecting a value in 'YYYY-MM-DDTHH:mm:ss' format. _PartioningAwareFileIndex_ considers these options during the process of checking for files, just before considering applied _PathFilters_ such as `pathGlobFilter.` In order to filter file results, a new PathFilter class was derived for this purpose. General house-keeping around classes extending PathFilter was performed for neatness. It became apparent support was needed to handle multiple potential path filters. Logic was introduced for this purpose and the associated tests written.
### Why are the changes needed?
When loading files from a data source, there can often times be thousands of file within a respective file path. In many cases I've seen, we want to start loading from a folder path and ideally be able to begin loading files having modification dates past a certain point. This would mean out of thousands of potential files, only the ones with modification dates greater than the specified timestamp would be considered. This saves a ton of time automatically and reduces significant complexity managing this in code.
### Does this PR introduce _any_ user-facing change?
This PR introduces an option that can be used with batch-based Spark file data sources. A documentation update was made to reflect an example and usage of the new data source option.
**Example Usages**
_Load all CSV files modified after date:_
`spark.read.format("csv").option("modifiedAfter","2020-06-15T05:00:00").load()`
_Load all CSV files modified before date:_
`spark.read.format("csv").option("modifiedBefore","2020-06-15T05:00:00").load()`
_Load all CSV files modified between two dates:_
`spark.read.format("csv").option("modifiedAfter","2019-01-15T05:00:00").option("modifiedBefore","2020-06-15T05:00:00").load()
`
### How was this patch tested?
A handful of unit tests were added to support the positive, negative, and edge case code paths.
It's also live in a handful of our Databricks dev environments. (quoted from cchighman)
Closes#30411 from HeartSaVioR/SPARK-31962.
Lead-authored-by: CC Highman <christopher.highman@microsoft.com>
Co-authored-by: Jungtaek Lim (HeartSaVioR) <kabhwan.opensource@gmail.com>
Signed-off-by: Jungtaek Lim (HeartSaVioR) <kabhwan.opensource@gmail.com>
### What changes were proposed in this pull request?
impl a new strategy `mode`: replace missing using the most frequent value along each column.
### Why are the changes needed?
it is highly scalable, and had been a function in [sklearn.impute.SimpleImputer](https://scikit-learn.org/stable/modules/generated/sklearn.impute.SimpleImputer.html#sklearn.impute.SimpleImputer) for a long time.
### Does this PR introduce _any_ user-facing change?
Yes, a new strategy is added
### How was this patch tested?
updated testsuites
Closes#30397 from zhengruifeng/imputer_max_freq.
Lead-authored-by: Ruifeng Zheng <ruifengz@foxmail.com>
Co-authored-by: zhengruifeng <ruifengz@foxmail.com>
Signed-off-by: Sean Owen <srowen@gmail.com>
### What changes were proposed in this pull request?
use `maxBlockSizeInMB` instead of `blockSize` (#rows) to control the stacking of vectors;
### Why are the changes needed?
the performance gain is mainly related to the nnz of block.
### Does this PR introduce _any_ user-facing change?
yes, param blockSize -> blockSizeInMB in master
### How was this patch tested?
updated testsuites
Closes#30355 from zhengruifeng/adaptively_blockify_aft_lir_lor.
Lead-authored-by: zhengruifeng <ruifengz@foxmail.com>
Co-authored-by: Ruifeng Zheng <ruifengz@foxmail.com>
Signed-off-by: Weichen Xu <weichen.xu@databricks.com>
### What changes were proposed in this pull request?
This change adds MapType support for PySpark with Arrow, if using pyarrow >= 2.0.0.
### Why are the changes needed?
MapType was previous unsupported with Arrow.
### Does this PR introduce _any_ user-facing change?
User can now enable MapType for `createDataFrame()`, `toPandas()` with Arrow optimization, and with Pandas UDFs.
### How was this patch tested?
Added new PySpark tests for createDataFrame(), toPandas() and Scalar Pandas UDFs.
Closes#30393 from BryanCutler/arrow-add-MapType-SPARK-24554.
Authored-by: Bryan Cutler <cutlerb@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
### What changes were proposed in this pull request?
This minor PR updates the docs of `schema_of_csv` and `schema_of_json`. They allow foldable string column instead of a string literal now.
### Why are the changes needed?
The function doc of `schema_of_csv` and `schema_of_json` are not updated accordingly with previous PRs.
### Does this PR introduce _any_ user-facing change?
Yes, update user-facing doc.
### How was this patch tested?
Unit test.
Closes#30396 from viirya/minor-json-csv.
Authored-by: Liang-Chi Hsieh <viirya@gmail.com>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>