#
# 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.
#
"""
Spark related features. Usually, the features here are missing in pandas
but Spark has it.
"""
from abc import ABCMeta, abstractmethod
from typing import TYPE_CHECKING, Callable, Generic, List, Optional, Union
from pyspark import StorageLevel
from pyspark.sql import Column as PySparkColumn, DataFrame as PySparkDataFrame
from pyspark.sql.types import DataType, StructType
from pyspark.pandas._typing import IndexOpsLike
from pyspark.pandas.internal import InternalField
if TYPE_CHECKING:
from pyspark.sql._typing import OptionalPrimitiveType
from pyspark._typing import PrimitiveType
import pyspark.pandas as ps
from pyspark.pandas.frame import CachedDataFrame
class SparkIndexOpsMethods(Generic[IndexOpsLike], metaclass=ABCMeta):
"""Spark related features. Usually, the features here are missing in pandas
but Spark has it."""
def __init__(self, data: IndexOpsLike):
self._data = data
@property
def data_type(self) -> DataType:
"""Returns the data type as defined by Spark, as a Spark DataType object."""
return self._data._internal.spark_type_for(self._data._column_label)
@property
def nullable(self) -> bool:
"""Returns the nullability as defined by Spark."""
return self._data._internal.spark_column_nullable_for(self._data._column_label)
@property
def column(self) -> PySparkColumn:
"""
Spark Column object representing the Series/Index.
.. note:: This Spark Column object is strictly stick to its base DataFrame the Series/Index
was derived from.
"""
return self._data._internal.spark_column_for(self._data._column_label)
def transform(self, func: Callable[[PySparkColumn], PySparkColumn]) -> IndexOpsLike:
"""
Applies a function that takes and returns a Spark column. It allows natively
applying a Spark function and column APIs with the Spark column internally used
in Series or Index. The output length of the Spark column should be the same as input's.
.. note:: It requires to have the same input and output length; therefore,
the aggregate Spark functions such as count does not work.
Parameters
----------
func : function
Function to use for transforming the data by using Spark columns.
Returns
-------
Series or Index
Raises
------
ValueError : If the output from the function is not a Spark column.
Examples
--------
>>> from pyspark.sql.functions import log
>>> df = ps.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]}, columns=["a", "b"])
>>> df
a b
0 1 4
1 2 5
2 3 6
>>> df.a.spark.transform(lambda c: log(c))
0 0.000000
1 0.693147
2 1.098612
Name: a, dtype: float64
>>> df.index.spark.transform(lambda c: c + 10)
Index([10, 11, 12], dtype='int64')
>>> df.a.spark.transform(lambda c: c + df.b.spark.column)
0 5
1 7
2 9
Name: a, dtype: int64
"""
from pyspark.pandas import MultiIndex
if isinstance(self._data, MultiIndex):
raise NotImplementedError("MultiIndex does not support spark.transform yet.")
output = func(self._data.spark.column)
if not isinstance(output, PySparkColumn):
raise ValueError(
"The output of the function [%s] should be of a "
"pyspark.sql.Column; however, got [%s]." % (func, type(output))
)
# Trigger the resolution so it throws an exception if anything does wrong
# within the function, for example,
# `df1.a.spark.transform(lambda _: F.col("non-existent"))`.
field = InternalField.from_struct_field(
self._data._internal.spark_frame.select(output).schema.fields[0]
)
return self._data._with_new_scol(scol=output, field=field)
@property
@abstractmethod
def analyzed(self) -> IndexOpsLike:
pass
class SparkSeriesMethods(SparkIndexOpsMethods["ps.Series"]):
def apply(self, func: Callable[[PySparkColumn], PySparkColumn]) -> "ps.Series":
"""
Applies a function that takes and returns a Spark column. It allows to natively
apply a Spark function and column APIs with the Spark column internally used
in Series or Index.
.. note:: It forces to lose the index and end up using the default index. It is
preferred to use :meth:`Series.spark.transform` or `:meth:`DataFrame.spark.apply`
with specifying the `index_col`.
.. note:: It does not require to have the same length of the input and output.
However, it requires to create a new DataFrame internally which will require
to set `compute.ops_on_diff_frames` to compute even with the same origin
DataFrame is expensive, whereas :meth:`Series.spark.transform` does not
require it.
Parameters
----------
func : function
Function to apply the function against the data by using Spark columns.
Returns
-------
Series
Raises
------
ValueError : If the output from the function is not a Spark column.
Examples
--------
>>> from pyspark import pandas as ps
>>> from pyspark.sql.functions import count, lit
>>> df = ps.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]}, columns=["a", "b"])
>>> df
a b
0 1 4
1 2 5
2 3 6
>>> df.a.spark.apply(lambda c: count(c))
0 3
Name: a, dtype: int64
>>> df.a.spark.apply(lambda c: c + df.b.spark.column)
0 5
1 7
2 9
Name: a, dtype: int64
"""
from pyspark.pandas.frame import DataFrame
from pyspark.pandas.series import Series, first_series
from pyspark.pandas.internal import HIDDEN_COLUMNS
output = func(self._data.spark.column)
if not isinstance(output, PySparkColumn):
raise ValueError(
"The output of the function [%s] should be of a "
"pyspark.sql.Column; however, got [%s]." % (func, type(output))
)
assert isinstance(self._data, Series)
sdf = self._data._internal.spark_frame.drop(*HIDDEN_COLUMNS).select(output)
# Lose index.
return first_series(DataFrame(sdf)).rename(self._data.name)
@property
def analyzed(self) -> "ps.Series":
"""
Returns a new Series with the analyzed Spark DataFrame.
After multiple operations, the underlying Spark plan could grow huge
and make the Spark planner take a long time to finish the planning.
This function is for the workaround to avoid it.
.. note:: After analyzing, operations between the analyzed Series and the original one
will **NOT** work without setting a config `compute.ops_on_diff_frames` to `True`.
Returns
-------
Series
Examples
--------
>>> ser = ps.Series([1, 2, 3])
>>> ser
0 1
1 2
2 3
dtype: int64
The analyzed one should return the same value.
>>> ser.spark.analyzed
0 1
1 2
2 3
dtype: int64
However, it won't work with the same anchor Series.
>>> with ps.option_context('compute.ops_on_diff_frames', False):
... ser + ser.spark.analyzed
Traceback (most recent call last):
...
ValueError: ... enable 'compute.ops_on_diff_frames' option.
>>> with ps.option_context('compute.ops_on_diff_frames', True):
... (ser + ser.spark.analyzed).sort_index()
0 2
1 4
2 6
dtype: int64
"""
from pyspark.pandas.frame import DataFrame
from pyspark.pandas.series import first_series
return first_series(DataFrame(self._data._internal.resolved_copy))
class SparkIndexMethods(SparkIndexOpsMethods["ps.Index"]):
@property
def analyzed(self) -> "ps.Index":
"""
Returns a new Index with the analyzed Spark DataFrame.
After multiple operations, the underlying Spark plan could grow huge
and make the Spark planner take a long time to finish the planning.
This function is for the workaround to avoid it.
.. note:: After analyzing, operations between the analyzed Series and the original one
will **NOT** work without setting a config `compute.ops_on_diff_frames` to `True`.
Returns
-------
Index
Examples
--------
>>> import pyspark.pandas as ps
>>> idx = ps.Index([1, 2, 3])
>>> idx
Index([1, 2, 3], dtype='int64')
The analyzed one should return the same value.
>>> idx.spark.analyzed
Index([1, 2, 3], dtype='int64')
However, it won't work with the same anchor Index.
>>> with ps.option_context('compute.ops_on_diff_frames', False):
... idx + idx.spark.analyzed
Traceback (most recent call last):
...
ValueError: ... enable 'compute.ops_on_diff_frames' option.
>>> with ps.option_context('compute.ops_on_diff_frames', True):
... (idx + idx.spark.analyzed).sort_values()
Index([2, 4, 6], dtype='int64')
"""
from pyspark.pandas.frame import DataFrame
return DataFrame(self._data._internal.resolved_copy).index
class SparkFrameMethods:
"""Spark related features. Usually, the features here are missing in pandas
but Spark has it."""
def __init__(self, frame: "ps.DataFrame"):
self._psdf = frame
def schema(self, index_col: Optional[Union[str, List[str]]] = None) -> StructType:
"""
Returns the underlying Spark schema.
Returns
-------
pyspark.sql.types.StructType
The underlying Spark schema.
Parameters
----------
index_col: str or list of str, optional, default: None
Column names to be used in Spark to represent pandas-on-Spark's index. The index name
in pandas-on-Spark is ignored. By default, the index is always lost.
Examples
--------
>>> df = ps.DataFrame({'a': list('abc'),
... 'b': list(range(1, 4)),
... 'c': np.arange(3, 6).astype('i1'),
... 'd': np.arange(4.0, 7.0, dtype='float64'),
... 'e': [True, False, True],
... 'f': pd.date_range('20130101', periods=3)},
... columns=['a', 'b', 'c', 'd', 'e', 'f'])
>>> df.spark.schema().simpleString()
'struct<a:string,b:bigint,c:tinyint,d:double,e:boolean,f:timestamp>'
>>> df.spark.schema(index_col='index').simpleString()
'struct<index:bigint,a:string,b:bigint,c:tinyint,d:double,e:boolean,f:timestamp>'
"""
return self.frame(index_col).schema
def print_schema(self, index_col: Optional[Union[str, List[str]]] = None) -> None:
"""
Prints out the underlying Spark schema in the tree format.
Parameters
----------
index_col: str or list of str, optional, default: None
Column names to be used in Spark to represent pandas-on-Spark's index. The index name
in pandas-on-Spark is ignored. By default, the index is always lost.
Returns
-------
None
Examples
--------
>>> df = ps.DataFrame({'a': list('abc'),
... 'b': list(range(1, 4)),
... 'c': np.arange(3, 6).astype('i1'),
... 'd': np.arange(4.0, 7.0, dtype='float64'),
... 'e': [True, False, True],
... 'f': pd.date_range('20130101', periods=3)},
... columns=['a', 'b', 'c', 'd', 'e', 'f'])
>>> df.spark.print_schema() # doctest: +NORMALIZE_WHITESPACE
root
|-- a: string (nullable = false)
|-- b: long (nullable = false)
|-- c: byte (nullable = false)
|-- d: double (nullable = false)
|-- e: boolean (nullable = false)
|-- f: timestamp (nullable = false)
>>> df.spark.print_schema(index_col='index') # doctest: +NORMALIZE_WHITESPACE
root
|-- index: long (nullable = false)
|-- a: string (nullable = false)
|-- b: long (nullable = false)
|-- c: byte (nullable = false)
|-- d: double (nullable = false)
|-- e: boolean (nullable = false)
|-- f: timestamp (nullable = false)
"""
self.frame(index_col).printSchema()
def frame(self, index_col: Optional[Union[str, List[str]]] = None) -> PySparkDataFrame:
"""
Return the current DataFrame as a Spark DataFrame. :meth:`DataFrame.spark.frame` is an
alias of :meth:`DataFrame.to_spark`.
Parameters
----------
index_col: str or list of str, optional, default: None
Column names to be used in Spark to represent pandas-on-Spark's index. The index name
in pandas-on-Spark is ignored. By default, the index is always lost.
See Also
--------
DataFrame.to_spark
DataFrame.pandas_api
DataFrame.spark.frame
Examples
--------
By default, this method loses the index as below.
>>> df = ps.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6], 'c': [7, 8, 9]})
>>> df.to_spark().show() # doctest: +NORMALIZE_WHITESPACE
+---+---+---+
| a| b| c|
+---+---+---+
| 1| 4| 7|
| 2| 5| 8|
| 3| 6| 9|
+---+---+---+
>>> df = ps.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6], 'c': [7, 8, 9]})
>>> df.spark.frame().show() # doctest: +NORMALIZE_WHITESPACE
+---+---+---+
| a| b| c|
+---+---+---+
| 1| 4| 7|
| 2| 5| 8|
| 3| 6| 9|
+---+---+---+
If `index_col` is set, it keeps the index column as specified.
>>> df.to_spark(index_col="index").show() # doctest: +NORMALIZE_WHITESPACE
+-----+---+---+---+
|index| a| b| c|
+-----+---+---+---+
| 0| 1| 4| 7|
| 1| 2| 5| 8|
| 2| 3| 6| 9|
+-----+---+---+---+
Keeping an index column is useful when you want to call some Spark APIs and
convert it back to pandas-on-Spark DataFrame without creating a default index, which
can affect performance.
>>> spark_df = df.to_spark(index_col="index")
>>> spark_df = spark_df.filter("a == 2")
>>> spark_df.pandas_api(index_col="index") # doctest: +NORMALIZE_WHITESPACE
a b c
index
1 2 5 8
In case of multi-index, specify a list to `index_col`.
>>> new_df = df.set_index("a", append=True)
>>> new_spark_df = new_df.to_spark(index_col=["index_1", "index_2"])
>>> new_spark_df.show() # doctest: +NORMALIZE_WHITESPACE
+-------+-------+---+---+
|index_1|index_2| b| c|
+-------+-------+---+---+
| 0| 1| 4| 7|
| 1| 2| 5| 8|
| 2| 3| 6| 9|
+-------+-------+---+---+
Can be converted back to pandas-on-Spark DataFrame.
>>> new_spark_df.pandas_api(
... index_col=["index_1", "index_2"]) # doctest: +NORMALIZE_WHITESPACE
b c
index_1 index_2
0 1 4 7
1 2 5 8
2 3 6 9
"""
from pyspark.pandas.utils import name_like_string
psdf = self._psdf
data_column_names = []
data_columns = []
for i, (label, spark_column, column_name) in enumerate(
zip(
psdf._internal.column_labels,
psdf._internal.data_spark_columns,
psdf._internal.data_spark_column_names,
)
):
name = str(i) if label is None else name_like_string(label)
data_column_names.append(name)
if column_name != name:
spark_column = spark_column.alias(name)
data_columns.append(spark_column)
if index_col is None:
return psdf._internal.spark_frame.select(data_columns)
else:
if isinstance(index_col, str):
index_col = [index_col]
old_index_scols = psdf._internal.index_spark_columns
if len(index_col) != len(old_index_scols):
raise ValueError(
"length of index columns is %s; however, the length of the given "
"'index_col' is %s." % (len(old_index_scols), len(index_col))
)
if any(col in data_column_names for col in index_col):
raise ValueError("'index_col' cannot be overlapped with other columns.")
new_index_scols = [
index_scol.alias(col) for index_scol, col in zip(old_index_scols, index_col)
]
return psdf._internal.spark_frame.select(new_index_scols + data_columns)
def cache(self) -> "CachedDataFrame":
"""
Yields and caches the current DataFrame.
The pandas-on-Spark DataFrame is yielded as a protected resource and its corresponding
data is cached which gets uncached after execution goes off the context.
If you want to specify the StorageLevel manually, use :meth:`DataFrame.spark.persist`
See Also
--------
DataFrame.spark.persist
Examples
--------
>>> df = ps.DataFrame([(.2, .3), (.0, .6), (.6, .0), (.2, .1)],
... columns=['dogs', 'cats'])
>>> df
dogs cats
0 0.2 0.3
1 0.0 0.6
2 0.6 0.0
3 0.2 0.1
>>> with df.spark.cache() as cached_df:
... print(cached_df.count())
...
dogs 4
cats 4
dtype: int64
>>> df = df.spark.cache()
>>> df.to_pandas().mean(axis=1)
0 0.25
1 0.30
2 0.30
3 0.15
dtype: float64
To uncache the dataframe, use `unpersist` function
>>> df.spark.unpersist()
"""
from pyspark.pandas.frame import CachedDataFrame
self._psdf._update_internal_frame(
self._psdf._internal.resolved_copy, check_same_anchor=False
)
return CachedDataFrame(self._psdf._internal)
def persist(
self, storage_level: StorageLevel = StorageLevel.MEMORY_AND_DISK
) -> "CachedDataFrame":
"""
Yields and caches the current DataFrame with a specific StorageLevel.
If a StorageLevel is not given, the `MEMORY_AND_DISK` level is used by default like PySpark.
The pandas-on-Spark DataFrame is yielded as a protected resource and its corresponding
data is cached which gets uncached after execution goes off the context.
See Also
--------
DataFrame.spark.cache
Examples
--------
>>> import pyspark
>>> df = ps.DataFrame([(.2, .3), (.0, .6), (.6, .0), (.2, .1)],
... columns=['dogs', 'cats'])
>>> df
dogs cats
0 0.2 0.3
1 0.0 0.6
2 0.6 0.0
3 0.2 0.1
Set the StorageLevel to `MEMORY_ONLY`.
>>> with df.spark.persist(pyspark.StorageLevel.MEMORY_ONLY) as cached_df:
... print(cached_df.spark.storage_level)
... print(cached_df.count())
...
Memory Serialized 1x Replicated
dogs 4
cats 4
dtype: int64
Set the StorageLevel to `DISK_ONLY`.
>>> with df.spark.persist(pyspark.StorageLevel.DISK_ONLY) as cached_df:
... print(cached_df.spark.storage_level)
... print(cached_df.count())
...
Disk Serialized 1x Replicated
dogs 4
cats 4
dtype: int64
If a StorageLevel is not given, it uses `MEMORY_AND_DISK` by default.
>>> with df.spark.persist() as cached_df:
... print(cached_df.spark.storage_level)
... print(cached_df.count())
...
Disk Memory Serialized 1x Replicated
dogs 4
cats 4
dtype: int64
>>> df = df.spark.persist()
>>> df.to_pandas().mean(axis=1)
0 0.25
1 0.30
2 0.30
3 0.15
dtype: float64
To uncache the dataframe, use `unpersist` function
>>> df.spark.unpersist()
"""
from pyspark.pandas.frame import CachedDataFrame
self._psdf._update_internal_frame(
self._psdf._internal.resolved_copy, check_same_anchor=False
)
return CachedDataFrame(self._psdf._internal, storage_level=storage_level)
def hint(self, name: str, *parameters: "PrimitiveType") -> "ps.DataFrame":
"""
Specifies some hint on the current DataFrame.
Parameters
----------
name : A name of the hint.
parameters : Optional parameters.
Returns
-------
ret : DataFrame with the hint.
See Also
--------
broadcast : Marks a DataFrame as small enough for use in broadcast joins.
Examples
--------
>>> df1 = ps.DataFrame({'lkey': ['foo', 'bar', 'baz', 'foo'],
... 'value': [1, 2, 3, 5]},
... columns=['lkey', 'value']).set_index('lkey')
>>> df2 = ps.DataFrame({'rkey': ['foo', 'bar', 'baz', 'foo'],
... 'value': [5, 6, 7, 8]},
... columns=['rkey', 'value']).set_index('rkey')
>>> merged = df1.merge(df2.spark.hint("broadcast"), left_index=True, right_index=True)
>>> merged.spark.explain() # doctest: +ELLIPSIS
== Physical Plan ==
...
...BroadcastHashJoin...
...
"""
from pyspark.pandas.frame import DataFrame
internal = self._psdf._internal.resolved_copy
return DataFrame(internal.with_new_sdf(internal.spark_frame.hint(name, *parameters)))
def to_table(
self,
name: str,
format: Optional[str] = None,
mode: str = "overwrite",
partition_cols: Optional[Union[str, List[str]]] = None,
index_col: Optional[Union[str, List[str]]] = None,
**options: "OptionalPrimitiveType",
) -> None:
"""
Write the DataFrame into a Spark table. :meth:`DataFrame.spark.to_table`
is an alias of :meth:`DataFrame.to_table`.
Parameters
----------
name : str, required
Table name in Spark.
format : string, optional
Specifies the output data source format. Some common ones are:
- 'delta'
- 'parquet'
- 'orc'
- 'json'
- 'csv'
mode : str {'append', 'overwrite', 'ignore', 'error', 'errorifexists'}, default
'overwrite'. Specifies the behavior of the save operation when the table exists
already.
- 'append': Append the new data to existing data.
- 'overwrite': Overwrite existing data.
- 'ignore': Silently ignore this operation if data already exists.
- 'error' or 'errorifexists': Throw an exception if data already exists.
partition_cols : str or list of str, optional, default None
Names of partitioning columns
index_col: str or list of str, optional, default: None
Column names to be used in Spark to represent pandas-on-Spark's index. The index name
in pandas-on-Spark is ignored. By default, the index is always lost.
options
Additional options passed directly to Spark.
Returns
-------
None
See Also
--------
read_table
DataFrame.spark.to_spark_io
DataFrame.to_parquet
Examples
--------
>>> df = ps.DataFrame(dict(
... date=list(pd.date_range('2012-1-1 12:00:00', periods=3, freq='ME')),
... country=['KR', 'US', 'JP'],
... code=[1, 2 ,3]), columns=['date', 'country', 'code'])
>>> df
date country code
0 2012-01-31 12:00:00 KR 1
1 2012-02-29 12:00:00 US 2
2 2012-03-31 12:00:00 JP 3
>>> df.to_table('%s.my_table' % db, partition_cols='date')
"""
if "options" in options and isinstance(options.get("options"), dict) and len(options) == 1:
options = options.get("options") # type: ignore[assignment]
self._psdf.spark.frame(index_col=index_col).write.saveAsTable(
name=name, format=format, mode=mode, partitionBy=partition_cols, **options
)
def to_spark_io(
self,
path: Optional[str] = None,
format: Optional[str] = None,
mode: str = "overwrite",
partition_cols: Optional[Union[str, List[str]]] = None,
index_col: Optional[Union[str, List[str]]] = None,
**options: "OptionalPrimitiveType",
) -> None:
"""Write the DataFrame out to a Spark data source.
Parameters
----------
path : string, optional
Path to the data source.
format : string, optional
Specifies the output data source format. Some common ones are:
- 'delta'
- 'parquet'
- 'orc'
- 'json'
- 'csv'
mode : str {'append', 'overwrite', 'ignore', 'error', 'errorifexists'}, default
'overwrite'. Specifies the behavior of the save operation when data already exists.
- 'append': Append the new data to existing data.
- 'overwrite': Overwrite existing data.
- 'ignore': Silently ignore this operation if data already exists.
- 'error' or 'errorifexists': Throw an exception if data already exists.
partition_cols : str or list of str, optional
Names of partitioning columns
index_col: str or list of str, optional, default: None
Column names to be used in Spark to represent pandas-on-Spark's index. The index name
in pandas-on-Spark is ignored. By default, the index is always lost.
options : dict
All other options passed directly into Spark's data source.
Returns
-------
None
See Also
--------
read_spark_io
DataFrame.to_delta
DataFrame.to_parquet
DataFrame.to_table
DataFrame.spark.to_spark_io
Examples
--------
>>> df = ps.DataFrame(dict(
... date=list(pd.date_range('2012-1-1 12:00:00', periods=3, freq='ME')),
... country=['KR', 'US', 'JP'],
... code=[1, 2 ,3]), columns=['date', 'country', 'code'])
>>> df
date country code
0 2012-01-31 12:00:00 KR 1
1 2012-02-29 12:00:00 US 2
2 2012-03-31 12:00:00 JP 3
>>> df.spark.to_spark_io(path='%s/to_spark_io/foo.json' % path, format='json')
"""
if "options" in options and isinstance(options.get("options"), dict) and len(options) == 1:
options = options.get("options") # type: ignore[assignment]
self._psdf.spark.frame(index_col=index_col).write.save(
path=path, format=format, mode=mode, partitionBy=partition_cols, **options
)
def explain(self, extended: Optional[bool] = None, mode: Optional[str] = None) -> None:
"""
Prints the underlying (logical and physical) Spark plans to the console for debugging
purpose.
Parameters
----------
extended : boolean, default ``False``.
If ``False``, prints only the physical plan.
mode : string, default ``None``.
The expected output format of plans.
Returns
-------
None
Examples
--------
>>> df = ps.DataFrame({'id': range(10)})
>>> df.spark.explain() # doctest: +ELLIPSIS
== Physical Plan ==
...
>>> df.spark.explain(True) # doctest: +ELLIPSIS
== Parsed Logical Plan ==
...
== Analyzed Logical Plan ==
...
== Optimized Logical Plan ==
...
== Physical Plan ==
...
>>> df.spark.explain("extended") # doctest: +ELLIPSIS
== Parsed Logical Plan ==
...
== Analyzed Logical Plan ==
...
== Optimized Logical Plan ==
...
== Physical Plan ==
...
>>> df.spark.explain(mode="extended") # doctest: +ELLIPSIS
== Parsed Logical Plan ==
...
== Analyzed Logical Plan ==
...
== Optimized Logical Plan ==
...
== Physical Plan ==
...
"""
self._psdf._internal.to_internal_spark_frame.explain(extended, mode)
def apply(
self,
func: Callable[[PySparkDataFrame], PySparkDataFrame],
index_col: Optional[Union[str, List[str]]] = None,
) -> "ps.DataFrame":
"""
Applies a function that takes and returns a Spark DataFrame. It allows natively
apply a Spark function and column APIs with the Spark column internally used
in Series or Index.
.. note:: set `index_col` and keep the column named as so in the output Spark
DataFrame to avoid using the default index to prevent performance penalty.
If you omit `index_col`, it will use default index which is potentially
expensive in general.
.. note:: it will lose column labels. This is a synonym of
``func(psdf.to_spark(index_col)).pandas_api(index_col)``.
Parameters
----------
func : function
Function to apply the function against the data by using Spark DataFrame.
Returns
-------
DataFrame
Raises
------
ValueError : If the output from the function is not a Spark DataFrame.
Examples
--------
>>> psdf = ps.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]}, columns=["a", "b"])
>>> psdf
a b
0 1 4
1 2 5
2 3 6
>>> psdf.spark.apply(
... lambda sdf: sdf.selectExpr("a + b as c", "index"), index_col="index")
... # doctest: +NORMALIZE_WHITESPACE
c
index
0 5
1 7
2 9
The case below ends up with using the default index, which should be avoided
if possible.
>>> psdf.spark.apply(lambda sdf: sdf.groupby("a").count().sort("a"))
a count
0 1 1
1 2 1
2 3 1
"""
output = func(self.frame(index_col))
if not isinstance(output, PySparkDataFrame):
raise ValueError(
"The output of the function [%s] should be of a "
"pyspark.sql.DataFrame; however, got [%s]." % (func, type(output))
)
return output.pandas_api(index_col)
def repartition(self, num_partitions: int) -> "ps.DataFrame":
"""
Returns a new DataFrame partitioned by the given partitioning expressions. The
resulting DataFrame is hash partitioned.
Parameters
----------
num_partitions : int
The target number of partitions.
Returns
-------
DataFrame
Examples
--------
>>> psdf = ps.DataFrame({"age": [5, 5, 2, 2],
... "name": ["Bob", "Bob", "Alice", "Alice"]}).set_index("age")
>>> psdf.sort_index() # doctest: +NORMALIZE_WHITESPACE
name
age
2 Alice
2 Alice
5 Bob
5 Bob
>>> new_psdf = psdf.spark.repartition(7)
>>> new_psdf.to_spark().rdd.getNumPartitions()
7
>>> new_psdf.sort_index() # doctest: +NORMALIZE_WHITESPACE
name
age
2 Alice
2 Alice
5 Bob
5 Bob
"""
from pyspark.pandas.frame import DataFrame
internal = self._psdf._internal.resolved_copy
repartitioned_sdf = internal.spark_frame.repartition(num_partitions)
return DataFrame(internal.with_new_sdf(repartitioned_sdf))
def coalesce(self, num_partitions: int) -> "ps.DataFrame":
"""
Returns a new DataFrame that has exactly `num_partitions` partitions.
.. note:: This operation results in a narrow dependency, e.g. if you go from 1000
partitions to 100 partitions, there will not be a shuffle, instead each of the 100 new
partitions will claim 10 of the current partitions. If a larger number of partitions is
requested, it will stay at the current number of partitions. However, if you're doing a
drastic coalesce, e.g. to num_partitions = 1, this may result in your computation taking
place on fewer nodes than you like (e.g. one node in the case of num_partitions = 1). To
avoid this, you can call repartition(). This will add a shuffle step, but means the
current upstream partitions will be executed in parallel (per whatever the current
partitioning is).
Parameters
----------
num_partitions : int
The target number of partitions.
Returns
-------
DataFrame
Examples
--------
>>> psdf = ps.DataFrame({"age": [5, 5, 2, 2],
... "name": ["Bob", "Bob", "Alice", "Alice"]}).set_index("age")
>>> psdf.sort_index() # doctest: +NORMALIZE_WHITESPACE
name
age
2 Alice
2 Alice
5 Bob
5 Bob
>>> new_psdf = psdf.spark.coalesce(1)
>>> new_psdf.to_spark().rdd.getNumPartitions()
1
>>> new_psdf.sort_index() # doctest: +NORMALIZE_WHITESPACE
name
age
2 Alice
2 Alice
5 Bob
5 Bob
"""
from pyspark.pandas.frame import DataFrame
internal = self._psdf._internal.resolved_copy
coalesced_sdf = internal.spark_frame.coalesce(num_partitions)
return DataFrame(internal.with_new_sdf(coalesced_sdf))
def checkpoint(self, eager: bool = True) -> "ps.DataFrame":
"""Returns a checkpointed version of this DataFrame.
Checkpointing can be used to truncate the logical plan of this DataFrame, which is
especially useful in iterative algorithms where the plan may grow exponentially. It will be
saved to files inside the checkpoint directory set with `SparkContext.setCheckpointDir`.
Parameters
----------
eager : bool
Whether to checkpoint this DataFrame immediately
Returns
-------
DataFrame
Examples
--------
>>> psdf = ps.DataFrame({"a": ["a", "b", "c"]})
>>> psdf
a
0 a
1 b
2 c
>>> new_psdf = psdf.spark.checkpoint() # doctest: +SKIP
>>> new_psdf # doctest: +SKIP
a
0 a
1 b
2 c
"""
from pyspark.pandas.frame import DataFrame
internal = self._psdf._internal.resolved_copy
checkpointed_sdf = internal.spark_frame.checkpoint(eager)
return DataFrame(internal.with_new_sdf(checkpointed_sdf))
def local_checkpoint(self, eager: bool = True) -> "ps.DataFrame":
"""Returns a locally checkpointed version of this DataFrame.
Checkpointing can be used to truncate the logical plan of this DataFrame, which is
especially useful in iterative algorithms where the plan may grow exponentially. Local
checkpoints are stored in the executors using the caching subsystem and therefore they are
not reliable.
Parameters
----------
eager : bool
Whether to locally checkpoint this DataFrame immediately
Returns
-------
DataFrame
Examples
--------
>>> psdf = ps.DataFrame({"a": ["a", "b", "c"]})
>>> psdf
a
0 a
1 b
2 c
>>> new_psdf = psdf.spark.local_checkpoint()
>>> new_psdf
a
0 a
1 b
2 c
"""
from pyspark.pandas.frame import DataFrame
internal = self._psdf._internal.resolved_copy
checkpointed_sdf = internal.spark_frame.localCheckpoint(eager)
return DataFrame(internal.with_new_sdf(checkpointed_sdf))
@property
def analyzed(self) -> "ps.DataFrame":
"""
Returns a new DataFrame with the analyzed Spark DataFrame.
After multiple operations, the underlying Spark plan could grow huge
and make the Spark planner take a long time to finish the planning.
This function is for the workaround to avoid it.
.. note:: After analysis, operations between the analyzed DataFrame and the original one
will **NOT** work without setting a config `compute.ops_on_diff_frames` to `True`.
Returns
-------
DataFrame
Examples
--------
>>> df = ps.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]}, columns=["a", "b"])
>>> df
a b
0 1 4
1 2 5
2 3 6
The analyzed one should return the same value.
>>> df.spark.analyzed
a b
0 1 4
1 2 5
2 3 6
However, it won't work with the same anchor Series.
>>> with ps.option_context('compute.ops_on_diff_frames', False):
... df + df.spark.analyzed
Traceback (most recent call last):
...
ValueError: ... enable 'compute.ops_on_diff_frames' option.
>>> with ps.option_context('compute.ops_on_diff_frames', True):
... (df + df.spark.analyzed).sort_index()
a b
0 2 8
1 4 10
2 6 12
"""
from pyspark.pandas.frame import DataFrame
return DataFrame(self._psdf._internal.resolved_copy)
class CachedSparkFrameMethods(SparkFrameMethods):
"""Spark related features for cached DataFrame. This is usually created via
`df.spark.cache()`."""
def __init__(self, frame: "CachedDataFrame"):
super().__init__(frame)
@property
def storage_level(self) -> StorageLevel:
"""
Return the storage level of this cache.
Examples
--------
>>> import pyspark.pandas as ps
>>> import pyspark
>>> df = ps.DataFrame([(.2, .3), (.0, .6), (.6, .0), (.2, .1)],
... columns=['dogs', 'cats'])
>>> df
dogs cats
0 0.2 0.3
1 0.0 0.6
2 0.6 0.0
3 0.2 0.1
>>> with df.spark.cache() as cached_df:
... print(cached_df.spark.storage_level)
...
Disk Memory Deserialized 1x Replicated
Set the StorageLevel to `MEMORY_ONLY`.
>>> with df.spark.persist(pyspark.StorageLevel.MEMORY_ONLY) as cached_df:
... print(cached_df.spark.storage_level)
...
Memory Serialized 1x Replicated
"""
return self._psdf._cached.storageLevel
def unpersist(self) -> None:
"""
The `unpersist` function is used to uncache the pandas-on-Spark DataFrame when it
is not used with the `with` statement.
Returns
-------
None
Examples
--------
>>> df = ps.DataFrame([(.2, .3), (.0, .6), (.6, .0), (.2, .1)],
... columns=['dogs', 'cats'])
>>> df = df.spark.cache()
To uncache the dataframe, use `unpersist` function
>>> df.spark.unpersist()
"""
if self._psdf._cached.is_cached:
self._psdf._cached.unpersist()
def _test() -> None:
import os
import doctest
import shutil
import sys
import tempfile
import uuid
import numpy
import pandas
from pyspark.sql import SparkSession
import pyspark.pandas.spark.accessors
os.chdir(os.environ["SPARK_HOME"])
globs = pyspark.pandas.spark.accessors.__dict__.copy()
globs["np"] = numpy
globs["pd"] = pandas
globs["ps"] = pyspark.pandas
spark = (
SparkSession.builder.master("local[4]")
.appName("pyspark.pandas.spark.accessors tests")
.getOrCreate()
)
db_name = "db%s" % str(uuid.uuid4()).replace("-", "")
spark.sql("CREATE DATABASE %s" % db_name)
globs["db"] = db_name
path = tempfile.mkdtemp()
globs["path"] = path
failure_count, test_count = doctest.testmod(
pyspark.pandas.spark.accessors,
globs=globs,
optionflags=doctest.ELLIPSIS | doctest.NORMALIZE_WHITESPACE,
)
shutil.rmtree(path, ignore_errors=True)
spark.sql("DROP DATABASE IF EXISTS %s CASCADE" % db_name)
spark.stop()
if failure_count:
sys.exit(-1)
if __name__ == "__main__":
_test()
