# 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.

""":py:class:`DataFrame` — lazy, chainable query representation.

A :py:class:`DataFrame` is a logical plan over one or more data sources.

Methods that reshape the plan (:py:meth:`DataFrame.select`,

:py:meth:`DataFrame.filter`, :py:meth:`DataFrame.aggregate`,

:py:meth:`DataFrame.sort`, :py:meth:`DataFrame.join`,

:py:meth:`DataFrame.limit`, the set-operation methods, ...) return a new

:py:class:`DataFrame` and do no work until a terminal method such as

:py:meth:`DataFrame.collect`, :py:meth:`DataFrame.to_pydict`,

:py:meth:`DataFrame.show`, or one of the ``write_*`` methods is called.

DataFrames are produced from a

:py:class:`~datafusion.context.SessionContext`, typically via

:py:meth:`~datafusion.context.SessionContext.sql`,

:py:meth:`~datafusion.context.SessionContext.read_csv`,

:py:meth:`~datafusion.context.SessionContext.read_parquet`, or

:py:meth:`~datafusion.context.SessionContext.from_pydict`.

Examples:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2, 3], "b": [10, 20, 30]})

>>> df.filter(col("a") > 1).select("b").to_pydict()

{'b': [20, 30]}

See :ref:`user_guide_concepts` in the online documentation for a high-level

overview of the execution model.

"""

from __future__ import annotations

import sys

import warnings

from collections.abc import AsyncIterator, Iterable, Iterator, Sequence

from typing import (

TYPE_CHECKING,

Any,

Literal,

cast,

overload,

)

if sys.version_info >= (3, 13):

from warnings import deprecated

else:

from typing_extensions import deprecated

from datafusion._internal import DataFrame as DataFrameInternal

from datafusion._internal import DataFrameWriteOptions as DataFrameWriteOptionsInternal

from datafusion._internal import InsertOp as InsertOpInternal

from datafusion._internal import ParquetColumnOptions as ParquetColumnOptionsInternal

from datafusion._internal import ParquetWriterOptions as ParquetWriterOptionsInternal

from datafusion.expr import (

Expr,

SortExpr,

SortKey,

_to_raw_expr,

ensure_expr,

ensure_expr_list,

expr_list_to_raw_expr_list,

sort_list_to_raw_sort_list,

)

from datafusion.plan import ExecutionPlan, LogicalPlan

from datafusion.record_batch import RecordBatch, RecordBatchStream

if TYPE_CHECKING:

import pathlib

from collections.abc import Callable

import pandas as pd

import polars as pl

import pyarrow as pa

from datafusion.catalog import Table

from enum import Enum

class ExplainFormat(Enum):

"""Output format for explain plans.

Controls how the query plan is rendered in :py:meth:`DataFrame.explain`.

"""

INDENT = "indent"

"""Default indented text format."""

TREE = "tree"

"""Tree-style visual format with box-drawing characters."""

PGJSON = "pgjson"

"""PostgreSQL-compatible JSON format for use with visualization tools."""

GRAPHVIZ = "graphviz"

"""Graphviz DOT format for graph rendering."""

class ExplainAnalyzeLevel(Enum):

"""Which metrics :py:meth:`DataFrame.explain` reports when ``analyze=True``."""

SUMMARY = "summary"

"""Common metrics for finding which operator is slow."""

DEV = "dev"

"""All metrics, including those for deep operator-level introspection."""

class ExplainMetricCategory(Enum):

"""Category of metric reported by :py:meth:`DataFrame.explain` with ``analyze``."""

ROWS = "rows"

"""Row counts, such as ``output_rows``."""

BYTES = "bytes"

"""Byte sizes, such as ``output_bytes``."""

TIMING = "timing"

"""Elapsed times, such as ``elapsed_compute``."""

UNCATEGORIZED = "uncategorized"

"""Metrics that declare no category."""

# excerpt from deltalake

# https://github.com/apache/datafusion-python/pull/981#discussion_r1905619163

class Compression(Enum):

"""Enum representing the available compression types for Parquet files."""

UNCOMPRESSED = "uncompressed"

SNAPPY = "snappy"

GZIP = "gzip"

BROTLI = "brotli"

LZ4 = "lz4"

# lzo is not implemented yet

# https://github.com/apache/arrow-rs/issues/6970

# LZO = "lzo" # noqa: ERA001

ZSTD = "zstd"

LZ4_RAW = "lz4_raw"

@classmethod

def from_str(cls: type[Compression], value: str) -> Compression:

"""Convert a string to a Compression enum value.

Args:

value: The string representation of the compression type.

Returns:

The Compression enum lowercase value.

Raises:

ValueError: If the string does not match any Compression enum value.

"""

try:

return cls(value.lower())

except ValueError as err:

valid_values = str([item.value for item in Compression])

error_msg = f"""

{value} is not a valid Compression.

Valid values are: {valid_values}

"""

raise ValueError(error_msg) from err

def get_default_level(self) -> int | None:

"""Get the default compression level for the compression type.

Returns:

The default compression level for the compression type.

"""

# GZIP, BROTLI default values from deltalake repo

# https://github.com/apache/datafusion-python/pull/981#discussion_r1905619163

# ZSTD default value from delta-rs

# https://github.com/apache/datafusion-python/pull/981#discussion_r1904789223

if self == Compression.GZIP:

return 6

if self == Compression.BROTLI:

return 1

if self == Compression.ZSTD:

return 4

return None

class ParquetWriterOptions:

"""Advanced parquet writer options.

Allows settings the writer options that apply to the entire file. Some options can

also be set on a column by column basis, with the field ``column_specific_options``

(see ``ParquetColumnOptions``).

"""

def __init__(

self,

data_pagesize_limit: int = 1024 * 1024,

write_batch_size: int = 1024,

writer_version: str = "1.0",

skip_arrow_metadata: bool = False,

compression: str | None = "zstd(3)",

compression_level: int | None = None,

dictionary_enabled: bool | None = True,

dictionary_page_size_limit: int = 1024 * 1024,

statistics_enabled: str | None = "page",

max_row_group_size: int = 1024 * 1024,

created_by: str = "datafusion-python",

column_index_truncate_length: int | None = 64,

statistics_truncate_length: int | None = None,

data_page_row_count_limit: int = 20_000,

encoding: str | None = None,

bloom_filter_on_write: bool = False,

bloom_filter_fpp: float | None = None,

bloom_filter_ndv: int | None = None,

allow_single_file_parallelism: bool = True,

maximum_parallel_row_group_writers: int = 1,

maximum_buffered_record_batches_per_stream: int = 2,

column_specific_options: dict[str, ParquetColumnOptions] | None = None,

) -> None:

"""Initialize the ParquetWriterOptions.

Args:

data_pagesize_limit: Sets best effort maximum size of data page in bytes.

write_batch_size: Sets write_batch_size in bytes.

writer_version: Sets parquet writer version. Valid values are ``1.0`` and

``2.0``.

skip_arrow_metadata: Skip encoding the embedded arrow metadata in the

KV_meta.

compression: Compression type to use. Default is ``zstd(3)``.

Available compression types are

- ``uncompressed``: No compression.

- ``snappy``: Snappy compression.

- ``gzip(n)``: Gzip compression with level n.

- ``brotli(n)``: Brotli compression with level n.

- ``lz4``: LZ4 compression.

- ``lz4_raw``: LZ4_RAW compression.

- ``zstd(n)``: Zstandard compression with level n.

compression_level: Compression level to set.

dictionary_enabled: Sets if dictionary encoding is enabled. If ``None``,

uses the default parquet writer setting.

dictionary_page_size_limit: Sets best effort maximum dictionary page size,

in bytes.

statistics_enabled: Sets if statistics are enabled for any column Valid

values are ``none``, ``chunk``, and ``page``. If ``None``, uses the

default parquet writer setting.

max_row_group_size: Target maximum number of rows in each row group

(defaults to 1M rows). Writing larger row groups requires more memory

to write, but can get better compression and be faster to read.

created_by: Sets "created by" property.

column_index_truncate_length: Sets column index truncate length.

statistics_truncate_length: Sets statistics truncate length. If ``None``,

uses the default parquet writer setting.

data_page_row_count_limit: Sets best effort maximum number of rows in a data

page.

encoding: Sets default encoding for any column. Valid values are ``plain``,

``plain_dictionary``, ``rle``, ``bit_packed``, ``delta_binary_packed``,

``delta_length_byte_array``, ``delta_byte_array``, ``rle_dictionary``,

and ``byte_stream_split``. If ``None``, uses the default parquet writer

setting.

bloom_filter_on_write: Write bloom filters for all columns when creating

parquet files.

bloom_filter_fpp: Sets bloom filter false positive probability. If ``None``,

uses the default parquet writer setting

bloom_filter_ndv: Sets bloom filter number of distinct values. If ``None``,

uses the default parquet writer setting.

allow_single_file_parallelism: Controls whether DataFusion will attempt to

speed up writing parquet files by serializing them in parallel. Each

column in each row group in each output file are serialized in parallel

leveraging a maximum possible core count of

``n_files * n_row_groups * n_columns``.

maximum_parallel_row_group_writers: By default parallel parquet writer is

tuned for minimum memory usage in a streaming execution plan. You may

see a performance benefit when writing large parquet files by increasing

``maximum_parallel_row_group_writers`` and

``maximum_buffered_record_batches_per_stream`` if your system has idle

cores and can tolerate additional memory usage. Boosting these values is

likely worthwhile when writing out already in-memory data, such as from

a cached data frame.

maximum_buffered_record_batches_per_stream: See

``maximum_parallel_row_group_writers``.

column_specific_options: Overrides options for specific columns. If a column

is not a part of this dictionary, it will use the parameters provided

here.

"""

self.data_pagesize_limit = data_pagesize_limit

self.write_batch_size = write_batch_size

self.writer_version = writer_version

self.skip_arrow_metadata = skip_arrow_metadata

if compression_level is not None:

self.compression = f"{compression}({compression_level})"

else:

self.compression = compression

self.dictionary_enabled = dictionary_enabled

self.dictionary_page_size_limit = dictionary_page_size_limit

self.statistics_enabled = statistics_enabled

self.max_row_group_size = max_row_group_size

self.created_by = created_by

self.column_index_truncate_length = column_index_truncate_length

self.statistics_truncate_length = statistics_truncate_length

self.data_page_row_count_limit = data_page_row_count_limit

self.encoding = encoding

self.bloom_filter_on_write = bloom_filter_on_write

self.bloom_filter_fpp = bloom_filter_fpp

self.bloom_filter_ndv = bloom_filter_ndv

self.allow_single_file_parallelism = allow_single_file_parallelism

self.maximum_parallel_row_group_writers = maximum_parallel_row_group_writers

self.maximum_buffered_record_batches_per_stream = (

maximum_buffered_record_batches_per_stream

)

self.column_specific_options = column_specific_options

class ParquetColumnOptions:

"""Parquet options for individual columns.

Contains the available options that can be applied for an individual Parquet column,

replacing the global options in ``ParquetWriterOptions``.

"""

def __init__(

self,

encoding: str | None = None,

dictionary_enabled: bool | None = None,

compression: str | None = None,

statistics_enabled: str | None = None,

bloom_filter_enabled: bool | None = None,

bloom_filter_fpp: float | None = None,

bloom_filter_ndv: int | None = None,

) -> None:

"""Initialize the ParquetColumnOptions.

Args:

encoding: Sets encoding for the column path. Valid values are: ``plain``,

``plain_dictionary``, ``rle``, ``bit_packed``, ``delta_binary_packed``,

``delta_length_byte_array``, ``delta_byte_array``, ``rle_dictionary``,

and ``byte_stream_split``. These values are not case-sensitive. If

``None``, uses the default parquet options

dictionary_enabled: Sets if dictionary encoding is enabled for the column

path. If `None`, uses the default parquet options

compression: Sets default parquet compression codec for the column path.

Valid values are ``uncompressed``, ``snappy``, ``gzip(level)``, ``lzo``,

``brotli(level)``, ``lz4``, ``zstd(level)``, and ``lz4_raw``. These

values are not case-sensitive. If ``None``, uses the default parquet

options.

statistics_enabled: Sets if statistics are enabled for the column Valid

values are: ``none``, ``chunk``, and ``page`` These values are not case

sensitive. If ``None``, uses the default parquet options.

bloom_filter_enabled: Sets if bloom filter is enabled for the column path.

If ``None``, uses the default parquet options.

bloom_filter_fpp: Sets bloom filter false positive probability for the

column path. If ``None``, uses the default parquet options.

bloom_filter_ndv: Sets bloom filter number of distinct values. If ``None``,

uses the default parquet options.

"""

self.encoding = encoding

self.dictionary_enabled = dictionary_enabled

self.compression = compression

self.statistics_enabled = statistics_enabled

self.bloom_filter_enabled = bloom_filter_enabled

self.bloom_filter_fpp = bloom_filter_fpp

self.bloom_filter_ndv = bloom_filter_ndv

class DataFrame:

"""Two dimensional table representation of data.

DataFrame objects are iterable; iterating over a DataFrame yields

:class:`datafusion.RecordBatch` instances lazily.

See :ref:`user_guide_concepts` in the online documentation for more information.

"""

def __init__(self, df: DataFrameInternal) -> None:

"""This constructor is not to be used by the end user.

See :py:class:`~datafusion.context.SessionContext` for methods to

create a :py:class:`DataFrame`.

"""

self.df = df

def into_view(self, temporary: bool = False) -> Table:

"""Convert ``DataFrame`` into a :class:`~datafusion.Table`.

Examples:

>>> from datafusion import SessionContext

>>> ctx = SessionContext()

>>> df = ctx.sql("SELECT 1 AS value")

>>> view = df.into_view()

>>> ctx.register_table("values_view", view)

>>> result = ctx.sql("SELECT value FROM values_view").collect()

>>> result[0].column("value").to_pylist()

[1]

"""

from datafusion.catalog import Table as _Table # noqa: PLC0415

return _Table(self.df.into_view(temporary))

def __getitem__(self, key: str | list[str]) -> DataFrame:

"""Return a new :py:class:`DataFrame` with the specified column or columns.

Args:

key: Column name or list of column names to select.

Returns:

DataFrame with the specified column or columns.

"""

return DataFrame(self.df.__getitem__(key))

def __repr__(self) -> str:

"""Return a string representation of the DataFrame.

Returns:

String representation of the DataFrame.

"""

return self.df.__repr__()

def _repr_html_(self) -> str:

return self.df._repr_html_()

@staticmethod

def default_str_repr(

batches: list[pa.RecordBatch],

schema: pa.Schema,

has_more: bool,

table_uuid: str | None = None,

) -> str:

"""Return the default string representation of a DataFrame.

This method is used by the default formatter and implemented in Rust for

performance reasons.

"""

return DataFrameInternal.default_str_repr(batches, schema, has_more, table_uuid)

def describe(self) -> DataFrame:

"""Return the statistics for this DataFrame.

Only summarized numeric datatypes at the moments and returns nulls

for non-numeric datatypes.

The output format is modeled after pandas.

Returns:

A summary DataFrame containing statistics.

"""

return DataFrame(self.df.describe())

def schema(self) -> pa.Schema:

"""Return the :py:class:`pyarrow.Schema` of this DataFrame.

The output schema contains information on the name, data type, and

nullability for each column.

Returns:

Describing schema of the DataFrame

"""

return self.df.schema()

def column(self, name: str) -> Expr:

"""Return a fully qualified column expression for ``name``.

Resolves an unqualified column name against this DataFrame's schema

and returns an :py:class:`Expr` whose underlying column reference

includes the table qualifier. This is especially useful after joins,

where the same column name may appear in multiple relations.

Args:

name: Unqualified column name to look up.

Returns:

A fully qualified column expression.

Raises:

Exception: If the column is not found or is ambiguous (exists in

multiple relations).

Examples:

Resolve a column from a simple DataFrame:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2], "b": [3, 4]})

>>> expr = df.column("a")

>>> df.select(expr).to_pydict()

{'a': [1, 2]}

Resolve qualified columns after a join:

>>> left = ctx.from_pydict({"id": [1, 2], "x": [10, 20]})

>>> right = ctx.from_pydict({"id": [1, 2], "y": [30, 40]})

>>> joined = left.join(right, on="id", how="inner")

>>> expr = joined.column("y")

>>> joined.select("id", expr).sort("id").to_pydict()

{'id': [1, 2], 'y': [30, 40]}

"""

return self.find_qualified_columns(name)[0]

def col(self, name: str) -> Expr:

"""Alias for :py:meth:`column`.

See Also:

:py:meth:`column`

"""

return self.column(name)

def find_qualified_columns(self, *names: str) -> list[Expr]:

"""Return fully qualified column expressions for the given names.

This is a batch version of :py:meth:`column` — it resolves each

unqualified name against the DataFrame's schema and returns a list

of qualified column expressions.

Args:

names: Unqualified column names to look up.

Returns:

List of fully qualified column expressions, one per name.

Raises:

Exception: If any column is not found or is ambiguous.

Examples:

Resolve multiple columns at once:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2], "b": [3, 4], "c": [5, 6]})

>>> exprs = df.find_qualified_columns("a", "c")

>>> df.select(*exprs).to_pydict()

{'a': [1, 2], 'c': [5, 6]}

"""

raw_exprs = self.df.find_qualified_columns(list(names))

return [Expr(e) for e in raw_exprs]

def select_exprs(self, *args: str) -> DataFrame:

"""Project arbitrary list of expression strings into a new DataFrame.

This method will parse string expressions into logical plan expressions.

The output DataFrame has one column for each expression.

Returns:

DataFrame only containing the specified columns.

"""

return self.df.select_exprs(*args)

def alias(self, alias: str) -> DataFrame:

"""Assign a table alias to this :py:class:`DataFrame`.

Replaces the qualifiers of the output columns with ``alias``. Useful for

self-joins and any situation that needs an unambiguous table-style

qualifier (``alias.col``) for downstream references.

Args:

alias: Table alias to apply to the DataFrame's columns.

Returns:

DataFrame with columns re-qualified under ``alias``.

Example:

>>> from datafusion import col

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"id": [1, 2], "val": [10, 20]})

>>> left = df.alias("l")

>>> right = df.alias("r")

>>> left.join(right, left_on="id", right_on="id").select(

... "id", col("l.val").alias("lval"), col("r.val").alias("rval")

... ).sort("id").to_pydict()

{'id': [1, 2], 'lval': [10, 20], 'rval': [10, 20]}

"""

return DataFrame(self.df.alias(alias))

def select(self, *exprs: Expr | str) -> DataFrame:

"""Project arbitrary expressions into a new :py:class:`DataFrame`.

String arguments are treated as column names; :py:class:`~datafusion.expr.Expr`

arguments can reshape, rename, or compute new columns.

Args:

exprs: Either column names or :py:class:`~datafusion.expr.Expr` to select.

Returns:

DataFrame after projection. It has one column for each expression.

Examples:

Select columns by name:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2, 3], "b": [10, 20, 30]})

>>> df.select("a").to_pydict()

{'a': [1, 2, 3]}

Mix column names, expressions, and aliases. The string ``"a"`` selects

column ``a`` directly; ``col("a").alias("alternate_a")`` returns a

duplicate under a new name:

>>> df.select("a", col("b"), col("a").alias("alternate_a")).to_pydict()

{'a': [1, 2, 3], 'b': [10, 20, 30], 'alternate_a': [1, 2, 3]}

"""

exprs_internal = expr_list_to_raw_expr_list(exprs)

return DataFrame(self.df.select(*exprs_internal))

def drop(self, *columns: str) -> DataFrame:

"""Drop arbitrary amount of columns.

Column names are case-sensitive and require double quotes to be dropped

if the original name is not strictly lower case.

Args:

columns: Column names to drop from the dataframe.

Returns:

DataFrame with those columns removed in the projection.

Examples:

To drop a lower-cased column 'a'

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2], "b": [3, 4]})

>>> df.drop("a").schema().names

['b']

Or to drop an upper-cased column 'A'

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"A": [1, 2], "b": [3, 4]})

>>> df.drop('"A"').schema().names

['b']

"""

return DataFrame(self.df.drop(*columns))

def window(self, *exprs: Expr) -> DataFrame:

"""Add window function columns to the DataFrame.

Applies the given window function expressions and appends the results

as new columns.

Args:

exprs: Window function expressions to evaluate.

Returns:

DataFrame with new window function columns appended.

Examples:

Add a row number within each group:

>>> import datafusion.functions as f

>>> from datafusion import col

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2, 3], "b": ["x", "x", "y"]})

>>> df = df.window(

... f.row_number(

... partition_by=[col("b")], order_by=[col("a")]

... ).alias("rn")

... )

>>> "rn" in df.schema().names

True

"""

raw = expr_list_to_raw_expr_list(exprs)

return DataFrame(self.df.window(*raw))

def filter(self, *predicates: Expr | str) -> DataFrame:

"""Return a DataFrame for which ``predicate`` evaluates to ``True``.

Rows for which ``predicate`` evaluates to ``False`` or ``None`` are filtered

out. If more than one predicate is provided, these predicates will be

combined as a logical AND. Each ``predicate`` can be an

:class:`~datafusion.expr.Expr` created using helper functions such as

:func:`datafusion.col` or :func:`datafusion.lit`, or a SQL expression string

that will be parsed against the DataFrame schema. If more complex logic is

required, see the logical operations in :py:mod:`~datafusion.functions`.

Examples:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2, 3]})

>>> df.filter(col("a") > lit(1)).to_pydict()

{'a': [2, 3]}

>>> df.filter("a > 1").to_pydict()

{'a': [2, 3]}

Args:

predicates: Predicate expression(s) or SQL strings to filter the DataFrame.

Returns:

DataFrame after filtering.

"""

df = self.df

for predicate in predicates:

expr = (

self.parse_sql_expr(predicate)

if isinstance(predicate, str)

else predicate

)

df = df.filter(ensure_expr(expr))

return DataFrame(df)

def parse_sql_expr(self, expr: str) -> Expr:

"""Creates logical expression from a SQL query text.

The expression is created and processed against the current schema.

Examples:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2, 3]})

>>> expr = df.parse_sql_expr("a > 1")

>>> df.filter(expr).to_pydict()

{'a': [2, 3]}

Args:

expr: Expression string to be converted to datafusion expression

Returns:

Logical expression .

"""

return Expr(self.df.parse_sql_expr(expr))

def with_column(self, name: str, expr: Expr | str) -> DataFrame:

"""Add an additional column to the DataFrame.

The ``expr`` must be an :class:`~datafusion.expr.Expr` constructed with

:func:`datafusion.col` or :func:`datafusion.lit`, or a SQL expression

string that will be parsed against the DataFrame schema.

Examples:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2]})

>>> df.with_column("b", col("a") + lit(10)).to_pydict()

{'a': [1, 2], 'b': [11, 12]}

Args:

name: Name of the column to add.

expr: Expression to compute the column.

Returns:

DataFrame with the new column.

"""

expr = self.parse_sql_expr(expr) if isinstance(expr, str) else expr

return DataFrame(self.df.with_column(name, ensure_expr(expr)))

def with_columns(

self, *exprs: Expr | str | Iterable[Expr | str], **named_exprs: Expr | str

) -> DataFrame:

"""Add columns to the DataFrame.

By passing expressions, iterables of expressions, string SQL expressions,

or named expressions.

All expressions must be :class:`~datafusion.expr.Expr` objects created via

:func:`datafusion.col` or :func:`datafusion.lit`, or SQL expression strings.

To pass named expressions use the form ``name=Expr``.

Example usage: The following will add 4 columns labeled ``a``, ``b``, ``c``,

and ``d``::

from datafusion import col, lit

df = df.with_columns(

col("x").alias("a"),

[lit(1).alias("b"), col("y").alias("c")],

d=lit(3)

)

Equivalent example using just SQL strings:

df = df.with_columns(

"x as a",

["1 as b", "y as c"],

d="3"

)

Args:

exprs: Either a single expression, an iterable of expressions to add or

SQL expression strings.

named_exprs: Named expressions in the form of ``name=expr``

Returns:

DataFrame with the new columns added.

"""

expressions = []

for expr in exprs:

if isinstance(expr, str):

expressions.append(self.parse_sql_expr(expr).expr)

elif isinstance(expr, Iterable) and not isinstance(

expr, Expr | str | bytes | bytearray

):

expressions.extend(

[

self.parse_sql_expr(e).expr

if isinstance(e, str)

else ensure_expr(e)

for e in expr

]

)

else:

expressions.append(ensure_expr(expr))

for alias, expr in named_exprs.items():

e = self.parse_sql_expr(expr) if isinstance(expr, str) else expr

ensure_expr(e)

expressions.append(e.alias(alias).expr)

return DataFrame(self.df.with_columns(expressions))

def with_column_renamed(self, old_name: str, new_name: str) -> DataFrame:

r"""Rename one column by applying a new projection.

This is a no-op if the column to be renamed does not exist.

The method supports case sensitive rename with wrapping column name

into one the following symbols (" or ' or \`).

Args:

old_name: Old column name.

new_name: New column name.

Returns:

DataFrame with the column renamed.

"""

return DataFrame(self.df.with_column_renamed(old_name, new_name))

def aggregate(

self,

group_by: Sequence[Expr | str] | Expr | str | None,

aggs: Sequence[Expr] | Expr,

) -> DataFrame:

"""Aggregates the rows of the current DataFrame.

By default each unique combination of the ``group_by`` columns

produces one row. To get multiple levels of subtotals in a

single pass, pass a

:py:class:`~datafusion.expr.GroupingSet` expression

(created via

:py:meth:`~datafusion.expr.GroupingSet.rollup`,

:py:meth:`~datafusion.expr.GroupingSet.cube`, or

:py:meth:`~datafusion.expr.GroupingSet.grouping_sets`)

as the ``group_by`` argument. See the

:ref:`aggregation` user guide for detailed examples.

Args:

group_by: Sequence of expressions or column names to group

by, or ``None`` for aggregation over the whole DataFrame.

A :py:class:`~datafusion.expr.GroupingSet` expression may

be included to produce multiple grouping levels (rollup,

cube, or explicit grouping sets).

aggs: Sequence of expressions to aggregate.

Returns:

DataFrame after aggregation.

Examples:

Aggregate without grouping — ``None`` or an empty ``group_by``

produces a single row:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict(

... {"team": ["x", "x", "y"], "score": [1, 2, 5]}

... )

>>> df.aggregate(None, [F.sum(col("score")).alias("total")]).to_pydict()

{'total': [8]}

Group by a column and produce one row per group:

>>> df.aggregate(

... ["team"], [F.sum(col("score")).alias("total")]

... ).sort("team").to_pydict()

{'team': ['x', 'y'], 'total': [3, 5]}

"""

if group_by is None:

group_by_list = []

else:

group_by_list = (

list(group_by)

if isinstance(group_by, Sequence)

and not isinstance(group_by, Expr | str)

else [group_by]

)

aggs_list = (

list(aggs)

if isinstance(aggs, Sequence) and not isinstance(aggs, Expr)

else [aggs]

)

group_by_exprs = expr_list_to_raw_expr_list(group_by_list)

aggs_exprs = ensure_expr_list(aggs_list)

return DataFrame(self.df.aggregate(group_by_exprs, aggs_exprs))

def sort(self, *exprs: SortKey) -> DataFrame:

"""Sort the DataFrame by the specified sorting expressions or column names.

Note that any expression can be turned into a sort expression by

calling its ``sort`` method. For ascending-only sorts, the shorter

:py:meth:`sort_by` is usually more convenient.

Args:

exprs: Sort expressions or column names, applied in order.

Returns:

DataFrame after sorting.

Examples:

Sort ascending by a column name:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [3, 1, 2], "b": [10, 20, 30]})

>>> df.sort("a").to_pydict()

{'a': [1, 2, 3], 'b': [20, 30, 10]}

Sort descending using :py:meth:`Expr.sort`:

>>> df.sort(col("a").sort(ascending=False)).to_pydict()

{'a': [3, 2, 1], 'b': [10, 30, 20]}

"""

exprs_raw = sort_list_to_raw_sort_list(exprs)

return DataFrame(self.df.sort(*exprs_raw))

def cast(self, mapping: dict[str, pa.DataType[Any]]) -> DataFrame:

"""Cast one or more columns to a different data type.

Args:

mapping: Mapped with column as key and column dtype as value.

Returns:

DataFrame after casting columns

"""

exprs = [Expr.column(col).cast(dtype) for col, dtype in mapping.items()]

return self.with_columns(exprs)

def limit(self, count: int, offset: int = 0) -> DataFrame:

"""Return a new :py:class:`DataFrame` with a limited number of rows.

Results are returned in unspecified order unless the DataFrame is

explicitly sorted first via :py:meth:`sort` or :py:meth:`sort_by`.

Args:

count: Number of rows to limit the DataFrame to.

offset: Number of rows to skip.

Returns:

DataFrame after limiting.

Examples:

Take the first two rows:

>>> ctx = dfn.SessionContext()

>>> df = ctx.from_pydict({"a": [1, 2, 3, 4]}).sort("a")

>>> df.limit(2).to_pydict()

{'a': [1, 2]}

Skip the first row then take two (paging):

>>> df.limit(2, offset=1).to_pydict()

{'a': [2, 3]}

"""

return DataFrame(self.df.limit(count, offset))

def head(self, n: int = 5) -> DataFrame:

"""Return a new :py:class:`DataFrame` with a limited number of rows.

Args:

n: Number of rows to take from the head of the DataFrame.

Returns:

DataFrame after limiting.

"""

return DataFrame(self.df.limit(n, 0))

def tail(self, n: int = 5) -> DataFrame:

"""Return a new :py:class:`DataFrame` with a limited number of rows.

Be aware this could be potentially expensive since the row size needs to be

determined of the dataframe. This is done by collecting it.

Args:

n: Number of rows to take from the tail of the DataFrame.

Returns:

DataFrame after limiting.

"""

return DataFrame(self.df.limit(n, max(0, self.count() - n)))

def collect(self) -> list[pa.RecordBatch]:

"""Execute this :py:class:`DataFrame` and collect results into memory.

Prior to calling ``collect``, modifying a DataFrame simply updates a plan

(no actual computation is performed). Calling ``collect`` triggers the

computation.

Returns:

List of :py:class:`pyarrow.RecordBatch` collected from the DataFrame.

"""

return self.df.collect()

def collect_column(self, column_name: str) -> pa.Array | pa.ChunkedArray:

"""Executes this :py:class:`DataFrame` for a single column."""