Coverage for pyspark/sql/pandas/group_ops.py : 45%

# 

# Licensed to the Apache Software Foundation (ASF) under one or more 

# contributor license agreements. See the NOTICE file distributed with 

# this work for additional information regarding copyright ownership. 

# The ASF licenses this file to You under the Apache License, Version 2.0 

# (the "License"); you may not use this file except in compliance with 

# the License. You may obtain a copy of the License at 

# 

# http://www.apache.org/licenses/LICENSE-2.0 

# 

# Unless required by applicable law or agreed to in writing, software 

# distributed under the License is distributed on an "AS IS" BASIS, 

# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 

# See the License for the specific language governing permissions and 

# limitations under the License. 

# 

import sys 

import warnings 

from pyspark.rdd import PythonEvalType 

from pyspark.sql.column import Column 

from pyspark.sql.dataframe import DataFrame 

class PandasGroupedOpsMixin(object): 

""" 

Min-in for pandas grouped operations. Currently, only :class:`GroupedData` 

can use this class. 

""" 

def apply(self, udf): 

""" 

It is an alias of :meth:`pyspark.sql.GroupedData.applyInPandas`; however, it takes a 

:meth:`pyspark.sql.functions.pandas_udf` whereas 

:meth:`pyspark.sql.GroupedData.applyInPandas` takes a Python native function. 

.. versionadded:: 2.3.0 

Parameters 

---------- 

udf : :func:`pyspark.sql.functions.pandas_udf` 

a grouped map user-defined function returned by 

:func:`pyspark.sql.functions.pandas_udf`. 

Notes 

----- 

It is preferred to use :meth:`pyspark.sql.GroupedData.applyInPandas` over this 

API. This API will be deprecated in the future releases. 

Examples 

-------- 

>>> from pyspark.sql.functions import pandas_udf, PandasUDFType 

>>> df = spark.createDataFrame( 

... [(1, 1.0), (1, 2.0), (2, 3.0), (2, 5.0), (2, 10.0)], 

... ("id", "v")) 

>>> @pandas_udf("id long, v double", PandasUDFType.GROUPED_MAP) # doctest: +SKIP 

... def normalize(pdf): 

... v = pdf.v 

... return pdf.assign(v=(v - v.mean()) / v.std()) 

>>> df.groupby("id").apply(normalize).show() # doctest: +SKIP 

+---+-------------------+ 

| id| v| 

+---+-------------------+ 

| 1|-0.7071067811865475| 

| 1| 0.7071067811865475| 

| 2|-0.8320502943378437| 

| 2|-0.2773500981126146| 

| 2| 1.1094003924504583| 

+---+-------------------+ 

See Also 

-------- 

pyspark.sql.functions.pandas_udf 

""" 

# Columns are special because hasattr always return True 

if isinstance(udf, Column) or not hasattr(udf, 'func') \ 

or udf.evalType != PythonEvalType.SQL_GROUPED_MAP_PANDAS_UDF: 

raise ValueError("Invalid udf: the udf argument must be a pandas_udf of type " 

"GROUPED_MAP.") 

warnings.warn( 

"It is preferred to use 'applyInPandas' over this " 

"API. This API will be deprecated in the future releases. See SPARK-28264 for " 

"more details.", UserWarning) 

return self.applyInPandas(udf.func, schema=udf.returnType) 

def applyInPandas(self, func, schema): 

""" 

Maps each group of the current :class:`DataFrame` using a pandas udf and returns the result 

as a `DataFrame`. 

The function should take a `pandas.DataFrame` and return another 

`pandas.DataFrame`. For each group, all columns are passed together as a `pandas.DataFrame` 

to the user-function and the returned `pandas.DataFrame` are combined as a 

:class:`DataFrame`. 

The `schema` should be a :class:`StructType` describing the schema of the returned 

`pandas.DataFrame`. The column labels of the returned `pandas.DataFrame` must either match 

the field names in the defined schema if specified as strings, or match the 

field data types by position if not strings, e.g. integer indices. 

The length of the returned `pandas.DataFrame` can be arbitrary. 

.. versionadded:: 3.0.0 

Parameters 

---------- 

func : function 

a Python native function that takes a `pandas.DataFrame`, and outputs a 

`pandas.DataFrame`. 

schema : :class:`pyspark.sql.types.DataType` or str 

the return type of the `func` in PySpark. The value can be either a 

:class:`pyspark.sql.types.DataType` object or a DDL-formatted type string. 

Examples 

-------- 

>>> import pandas as pd # doctest: +SKIP 

>>> from pyspark.sql.functions import pandas_udf, ceil 

>>> df = spark.createDataFrame( 

... [(1, 1.0), (1, 2.0), (2, 3.0), (2, 5.0), (2, 10.0)], 

... ("id", "v")) # doctest: +SKIP 

>>> def normalize(pdf): 

... v = pdf.v 

... return pdf.assign(v=(v - v.mean()) / v.std()) 

>>> df.groupby("id").applyInPandas( 

... normalize, schema="id long, v double").show() # doctest: +SKIP 

+---+-------------------+ 

| id| v| 

+---+-------------------+ 

| 1|-0.7071067811865475| 

| 1| 0.7071067811865475| 

| 2|-0.8320502943378437| 

| 2|-0.2773500981126146| 

| 2| 1.1094003924504583| 

+---+-------------------+ 

Alternatively, the user can pass a function that takes two arguments. 

In this case, the grouping key(s) will be passed as the first argument and the data will 

be passed as the second argument. The grouping key(s) will be passed as a tuple of numpy 

data types, e.g., `numpy.int32` and `numpy.float64`. The data will still be passed in 

as a `pandas.DataFrame` containing all columns from the original Spark DataFrame. 

This is useful when the user does not want to hardcode grouping key(s) in the function. 

>>> df = spark.createDataFrame( 

... [(1, 1.0), (1, 2.0), (2, 3.0), (2, 5.0), (2, 10.0)], 

... ("id", "v")) # doctest: +SKIP 

>>> def mean_func(key, pdf): 

... # key is a tuple of one numpy.int64, which is the value 

... # of 'id' for the current group 

... return pd.DataFrame([key + (pdf.v.mean(),)]) 

>>> df.groupby('id').applyInPandas( 

... mean_func, schema="id long, v double").show() # doctest: +SKIP 

+---+---+ 

| id| v| 

+---+---+ 

| 1|1.5| 

| 2|6.0| 

+---+---+ 

>>> def sum_func(key, pdf): 

... # key is a tuple of two numpy.int64s, which is the values 

... # of 'id' and 'ceil(df.v / 2)' for the current group 

... return pd.DataFrame([key + (pdf.v.sum(),)]) 

>>> df.groupby(df.id, ceil(df.v / 2)).applyInPandas( 

... sum_func, schema="id long, `ceil(v / 2)` long, v double").show() # doctest: +SKIP 

+---+-----------+----+ 

| id|ceil(v / 2)| v| 

+---+-----------+----+ 

| 2| 5|10.0| 

| 1| 1| 3.0| 

| 2| 3| 5.0| 

| 2| 2| 3.0| 

+---+-----------+----+ 

Notes 

----- 

This function requires a full shuffle. All the data of a group will be loaded 

into memory, so the user should be aware of the potential OOM risk if data is skewed 

and certain groups are too large to fit in memory. 

If returning a new `pandas.DataFrame` constructed with a dictionary, it is 

recommended to explicitly index the columns by name to ensure the positions are correct, 

or alternatively use an `OrderedDict`. 

For example, `pd.DataFrame({'id': ids, 'a': data}, columns=['id', 'a'])` or 

`pd.DataFrame(OrderedDict([('id', ids), ('a', data)]))`. 

This API is experimental. 

See Also 

-------- 

pyspark.sql.functions.pandas_udf 

""" 

from pyspark.sql import GroupedData 

from pyspark.sql.functions import pandas_udf, PandasUDFType 

assert isinstance(self, GroupedData) 

udf = pandas_udf( 

func, returnType=schema, functionType=PandasUDFType.GROUPED_MAP) 

df = self._df 

udf_column = udf(*[df[col] for col in df.columns]) 

jdf = self._jgd.flatMapGroupsInPandas(udf_column._jc.expr()) 

return DataFrame(jdf, self.sql_ctx) 

def cogroup(self, other): 

""" 

Cogroups this group with another group so that we can run cogrouped operations. 

.. versionadded:: 3.0.0 

See :class:`PandasCogroupedOps` for the operations that can be run. 

""" 

from pyspark.sql import GroupedData 

assert isinstance(self, GroupedData) 

return PandasCogroupedOps(self, other) 

class PandasCogroupedOps(object): 

""" 

A logical grouping of two :class:`GroupedData`, 

created by :func:`GroupedData.cogroup`. 

.. versionadded:: 3.0.0 

Notes 

----- 

This API is experimental. 

""" 

def __init__(self, gd1, gd2): 

self._gd1 = gd1 

self._gd2 = gd2 

self.sql_ctx = gd1.sql_ctx 

def applyInPandas(self, func, schema): 

""" 

Applies a function to each cogroup using pandas and returns the result 

as a `DataFrame`. 

The function should take two `pandas.DataFrame`\\s and return another 

`pandas.DataFrame`. For each side of the cogroup, all columns are passed together as a 

`pandas.DataFrame` to the user-function and the returned `pandas.DataFrame` are combined as 

a :class:`DataFrame`. 

The `schema` should be a :class:`StructType` describing the schema of the returned 

`pandas.DataFrame`. The column labels of the returned `pandas.DataFrame` must either match 

the field names in the defined schema if specified as strings, or match the 

field data types by position if not strings, e.g. integer indices. 

The length of the returned `pandas.DataFrame` can be arbitrary. 

.. versionadded:: 3.0.0 

Parameters 

---------- 

func : function 

a Python native function that takes two `pandas.DataFrame`\\s, and 

outputs a `pandas.DataFrame`, or that takes one tuple (grouping keys) and two 

pandas ``DataFrame``\\s, and outputs a pandas ``DataFrame``. 

schema : :class:`pyspark.sql.types.DataType` or str 

the return type of the `func` in PySpark. The value can be either a 

:class:`pyspark.sql.types.DataType` object or a DDL-formatted type string. 

Examples 

-------- 

>>> from pyspark.sql.functions import pandas_udf 

>>> df1 = spark.createDataFrame( 

... [(20000101, 1, 1.0), (20000101, 2, 2.0), (20000102, 1, 3.0), (20000102, 2, 4.0)], 

... ("time", "id", "v1")) 

>>> df2 = spark.createDataFrame( 

... [(20000101, 1, "x"), (20000101, 2, "y")], 

... ("time", "id", "v2")) 

>>> def asof_join(l, r): 

... return pd.merge_asof(l, r, on="time", by="id") 

>>> df1.groupby("id").cogroup(df2.groupby("id")).applyInPandas( 

... asof_join, schema="time int, id int, v1 double, v2 string" 

... ).show() # doctest: +SKIP 

+--------+---+---+---+ 

| time| id| v1| v2| 

+--------+---+---+---+ 

|20000101| 1|1.0| x| 

|20000102| 1|3.0| x| 

|20000101| 2|2.0| y| 

|20000102| 2|4.0| y| 

+--------+---+---+---+ 

Alternatively, the user can define a function that takes three arguments. In this case, 

the grouping key(s) will be passed as the first argument and the data will be passed as the 

second and third arguments. The grouping key(s) will be passed as a tuple of numpy data 

types, e.g., `numpy.int32` and `numpy.float64`. The data will still be passed in as two 

`pandas.DataFrame` containing all columns from the original Spark DataFrames. 

>>> def asof_join(k, l, r): 

... if k == (1,): 

... return pd.merge_asof(l, r, on="time", by="id") 

... else: 

... return pd.DataFrame(columns=['time', 'id', 'v1', 'v2']) 

>>> df1.groupby("id").cogroup(df2.groupby("id")).applyInPandas( 

... asof_join, "time int, id int, v1 double, v2 string").show() # doctest: +SKIP 

+--------+---+---+---+ 

| time| id| v1| v2| 

+--------+---+---+---+ 

|20000101| 1|1.0| x| 

|20000102| 1|3.0| x| 

+--------+---+---+---+ 

Notes 

----- 

This function requires a full shuffle. All the data of a cogroup will be loaded 

into memory, so the user should be aware of the potential OOM risk if data is skewed 

and certain groups are too large to fit in memory. 

If returning a new `pandas.DataFrame` constructed with a dictionary, it is 

recommended to explicitly index the columns by name to ensure the positions are correct, 

or alternatively use an `OrderedDict`. 

For example, `pd.DataFrame({'id': ids, 'a': data}, columns=['id', 'a'])` or 

`pd.DataFrame(OrderedDict([('id', ids), ('a', data)]))`. 

This API is experimental. 

See Also 

-------- 

pyspark.sql.functions.pandas_udf 

""" 

from pyspark.sql.pandas.functions import pandas_udf 

udf = pandas_udf( 

func, returnType=schema, functionType=PythonEvalType.SQL_COGROUPED_MAP_PANDAS_UDF) 

all_cols = self._extract_cols(self._gd1) + self._extract_cols(self._gd2) 

udf_column = udf(*all_cols) 

jdf = self._gd1._jgd.flatMapCoGroupsInPandas(self._gd2._jgd, udf_column._jc.expr()) 

return DataFrame(jdf, self.sql_ctx) 

@staticmethod 

def _extract_cols(gd): 

df = gd._df 

return [df[col] for col in df.columns] 

def _test(): 

import doctest 

from pyspark.sql import SparkSession 

import pyspark.sql.pandas.group_ops 

globs = pyspark.sql.pandas.group_ops.__dict__.copy() 

spark = SparkSession.builder\ 

.master("local[4]")\ 

.appName("sql.pandas.group tests")\ 

.getOrCreate() 

globs['spark'] = spark 

(failure_count, test_count) = doctest.testmod( 

pyspark.sql.pandas.group_ops, globs=globs, 

optionflags=doctest.ELLIPSIS | doctest.NORMALIZE_WHITESPACE | doctest.REPORT_NDIFF) 

spark.stop() 

355 ↛ 356line 355 didn't jump to line 356, because the condition on line 355 was never true if failure_count: 

sys.exit(-1) 

if __name__ == "__main__": 

_test()