ENH: DataFrameGroupBy.transform now accepts list, dict, and NamedAgg arguments (GH#58318) by berasaikat · Pull Request #65164 · pandas-dev/pandas

berasaikat · 2026-04-11T15:34:58Z

Description

What this does

GroupBy.transform currently only accepts a single callable or string, while GroupBy.agg supports list-like, dict, and NamedAgg keyword arguments. This PR closes that gap by extending DataFrameGroupBy.transform to support the same call signatures as GroupBy.agg.

df = pd.DataFrame({"col": list("aab"), "val": range(3), "other": range(3)})

# Before — TypeError: 'list' object is not callable
df.groupby("col").transform(["sum", "min"])

# Now works
df.groupby("col").transform(["sum", "min"])
df.groupby("col").transform({"val": "sum", "other": "min"})
df.groupby("col").transform(
    val_sum=pd.NamedAgg(column="val", aggfunc="sum"),
    other_min=pd.NamedAgg(column="other", aggfunc="min"),
)

Changes

pandas/core/groupby/generic.py

func is now optional (defaults to None) to support the NamedAgg kwargs path
List-like dispatch: applies every function to every non-key column, returning a MultiIndex-column DataFrame (column, func)
Dict dispatch: applies a per-column function, returning only the specified columns. Plain {"col": "sum"} and {"name": NamedAgg(column, func)} are both supported, as are plain tuples as a NamedAgg equivalent
NamedAgg / plain tuple kwargs dispatch: named output columns with explicit source column selection
Two new private helpers: _transform_multiple_funcs (orchestrates list and dict paths) and _transform_single_column (applies one function to one column via SeriesGroupBy)
NotImplementedError with a clear message for SeriesGroupBy.transform(list) and dict-of-lists — both are intended for a future release
ValueError for duplicate column names in dict/NamedAgg paths

pandas/tests/groupby/transform/test_transform.py

test_transform_with_list_like
test_transform_with_list_like_single_column
test_transform_with_dict
test_transform_with_dict_subset_columns
test_transform_with_dict_of_lists_raises
test_transform_with_namedagg
test_transform_with_namedagg_same_source_column
test_transform_with_namedagg_plain_tuple
test_transform_series_groupby_list_raises
test_transform_dict_duplicate_column_names_raises

doc/source/whatsnew/v3.1.0.rst

- :meth:`DataFrameGroupBy.transform` now accepts list-like and dict arguments
  similar to :meth:`GroupBy.agg`, and supports :class:`NamedAgg` (:issue:`58318`)

Checklist

Tests added and passed
Type annotations added
Entry added to doc/source/whatsnew/v3.1.0.rst

…arguments (GH#58318)

rhshadrach · 2026-04-23T20:53:30Z

Thanks for working on this, will take a look. If I don't update in a week, please send me a ping!

berasaikat · 2026-05-04T18:55:21Z

@rhshadrach - Let me know if anything needs to be updated!

rhshadrach

Looks very good, some comments. Regarding using NamedAgg, we'll need to make a decision here as to the naming since this isn't great when used with e.g. cumsum. The implementation doesn't limit it to aggregation functions already. I'd support adding an alias NamedFunc and the (possibly) deprecating NamedAgg.

cc @jbrockmendel

rhshadrach · 2026-05-05T21:15:52Z

+            raise NotImplementedError(
+                "Passing a list to SeriesGroupBy.transform is not yet supported "
+                "and is intended to be implemented in a future release. "
+                "See GH#58318."


For references within the pandas code this would be fine, but for user-facing I think we should have the full URL.

rhshadrach · 2026-05-05T21:17:17Z

+        1    1   0     1   0
+        2    2   2     2   2
+
+        .. versionchanged:: 3.0.0


3.1.0 throughout.

rhshadrach · 2026-05-05T21:18:29Z

+        if func is None:
+            # Named-aggregation style:
+            #   .transform(val_sum=NamedAgg(column="val", aggfunc="sum"), ...)
+            transformed_func: dict = dict(kwargs.items())


Why is this necessary?

rhshadrach · 2026-05-05T21:19:56Z

+                func, *args, engine=engine, engine_kwargs=engine_kwargs, **kwargs
+            )
+        else:
+            # Original single-function path; unchanged.


This comment will be not meaningful when this PR is merged; can you remove.

rhshadrach · 2026-05-05T21:21:00Z

+            return self._transform_multiple_funcs(
+                transformed_func, *args, engine=engine, engine_kwargs=engine_kwargs
+            )
+        elif isinstance(func, dict):


This is False for e.g. UserDict. Can you use is_dict_like.

rhshadrach · 2026-05-05T21:21:13Z

+            return self._transform_multiple_funcs(
+                func, *args, engine=engine, engine_kwargs=engine_kwargs, **kwargs
+            )
+        elif isinstance(func, list):


Similar - use is_list_like.

rhshadrach · 2026-05-05T21:21:22Z

+        """
+        from pandas.core.reshape.concat import concat
+
+        if isinstance(func, dict):


Same as above.

rhshadrach · 2026-05-05T21:23:04Z

+        # ── list path ────────────────────────────────────────────────────────
+        # Apply every func to every non-key column.
+        # _obj_with_exclusions already omits groupby keys and excluded columns.
+        assert isinstance(func, list)


Same as above.

rhshadrach · 2026-05-05T21:27:00Z

+        if (self._obj_with_exclusions.columns == column_name).sum() > 1:
+            raise ValueError(
+                f"Column label '{column_name}' is not unique in this DataFrame. "
+                "DataFrameGroupBy.transform with a dict or NamedAgg does not "
+                "support duplicate column names. See GH#58318."
+            )


Why do we need to impose this restriction?

berasaikat · 2026-05-10T12:10:37Z

@rhshadrach - addressed all the code comments. Thank you!

Added pd.NamedFunc as an alias for pd.NamedAgg per reviewer suggestion. Deprecation of NamedAgg is left as a separate follow-up.

jbrockmendel · 2026-05-11T21:20:07Z

I'd support adding an alias NamedFunc and the (possibly) deprecating NamedAgg.

Fine by me, no strong opinion.

rhshadrach

Looking good! Just some minor style requests.

rhshadrach · 2026-05-18T20:48:05Z

+        1    1   0     1   0
+        2    2   2     2   2
+
+        .. versionchanged:: 3.0.1


Instead of these versionchanged throughout the example, can you do a single one after the func argument description.

rhshadrach · 2026-05-18T20:49:59Z

+            ``output_name=NamedFunc(column, func)`` for named transformation or
+            ``output_name=NamedAgg(column, aggfunc)`` for named aggregation.


This makes it seem like you have to use NamedAgg if you have an aggregation function, which is not the case. I'd suggest just leaving NamedAgg off here.

Suggested change

``output_name=NamedFunc(column, func)`` for named transformation or

``output_name=NamedAgg(column, aggfunc)`` for named aggregation.

``output_name=NamedFunc(column, func)``.

rhshadrach · 2026-05-18T20:51:28Z

+        from pandas.core.reshape.concat import concat
+
+        if is_dict_like(func):
+            # ── dict / NamedAgg path ─────────────────────────────────────────


Suggested change

# ── dict / NamedAgg path ─────────────────────────────────────────

# Also includes NamedAgg / NamedFunc

rhshadrach · 2026-05-18T20:58:29Z

+                results.append(result)
+            return concat(results, axis=1)
+
+        # ── list path ────────────────────────────────────────────────────────


Suggested change

# ── list path ────────────────────────────────────────────────────────

# list path

rhshadrach · 2026-05-18T21:01:28Z

+
+        # ── list path ────────────────────────────────────────────────────────
+        # Apply every func to every non-key column.
+        # _obj_with_exclusions already omits groupby keys and excluded columns.


Suggested change

# _obj_with_exclusions already omits groupby keys and excluded columns.

I don't think this comment is adding anything.

berasaikat · 2026-05-20T19:26:29Z

Made the changes.

rhshadrach

I think we should just refer to NamedFunc when it comes to transform; also can you add one more note in the whatsnew for NamedFunc. Something along the lines of:

- Added :class:`NamedFunc`, an alias to :class:`NamedAgg` for a more general name; either can accept arbitrary functions (:issue:`65164`)

rhshadrach · 2026-05-28T09:44:37Z

            - Numba JIT function with ``engine='numba'`` specified.
+            - List of strings/functions: applied to every non-key column,
+              returning a MultiIndex-column DataFrame ``(column, func)``.
+            - Dict ``{column: func}`` or ``{name: NamedAgg(column, func)}``:


NamedFunc here

rhshadrach · 2026-05-28T09:45:26Z

              applied per-column as specified.

+            .. versionchanged:: 3.1.0
+                    Added support for list-like, dict, and :class:`NamedAgg` arguments.


NamedFunc here.

rhshadrach · 2026-05-28T09:46:09Z

 ^^^^^^^^^^^^^^^^^^
 - :class:`Period` now supports f-string formatting via ``__format__``, e.g. ``f"{period:%Y-%m}"`` (:issue:`48536`)
 - :meth:`.DataFrameGroupBy.agg` now allows for the provided ``func`` to return a NumPy array (:issue:`63957`)
+- :meth:`DataFrameGroupBy.transform` now accepts list-like and dict arguments similar to :meth:`GroupBy.agg`, and supports :class:`NamedAgg` (:issue:`58318`)


NamedFunc here

rhshadrach · 2026-05-28T09:52:16Z

    "MultiIndex",
    "NaT",
    "NamedAgg",
+    "NamedFunc",


This should also be imported in pandas/__init__.py. There will likely be some API tests that break; those tests just then need updated with what to expect.

Done. Minimal updates needed in API tests.

…es#58318

rhshadrach

lgtm

jbrockmendel · 2026-06-08T21:14:07Z

I don’t have bandwidth to do a proper review. No objections based on a quick glance. Happy to defer to rhshadrach

rhshadrach · 2026-06-09T22:12:07Z

Thanks @berasaikat

…arguments (GH#58318) (pandas-dev#65164)

ENH: DataFrameGroupBy.transform now accepts list, dict, and NamedAgg …

3d3a444

…arguments (GH#58318)

berasaikat requested a review from rhshadrach as a code owner April 11, 2026 15:35

berasaikat and others added 3 commits April 11, 2026 22:02

Fixed Doc Build Error

e0e314c

Merge branch 'main' into issues#58318

6728335

Trigger CI checks

cb5d000

Merge branch 'main' into issues#58318

fc7b1f2

rhshadrach requested changes May 5, 2026

View reviewed changes

berasaikat added 2 commits May 10, 2026 16:12

adressing review comments

0564635

fix typecheck

a3e5d2b

berasaikat requested a review from rhshadrach May 12, 2026 20:42

rhshadrach requested changes May 18, 2026

View reviewed changes

berasaikat added 3 commits May 20, 2026 02:50

style changes

1094348

fix doc error

e2f1c72

fix doc error 2

53c4d45

berasaikat requested a review from rhshadrach May 20, 2026 19:26

Trigger CI pipeline

e229b33

rhshadrach requested changes May 28, 2026

View reviewed changes

berasaikat added 2 commits June 5, 2026 02:25

Merge branch 'main' of https://github.com/berasaikat/pandas into issu…

79f0243

…es#58318

NamedFunc fix

88b713b

berasaikat requested a review from rhshadrach June 5, 2026 05:06

rhshadrach approved these changes Jun 8, 2026

View reviewed changes

rhshadrach requested a review from jbrockmendel June 8, 2026 20:21

rhshadrach merged commit c79d638 into pandas-dev:main Jun 9, 2026
46 checks passed

rhshadrach added the Groupby label Jun 9, 2026

rhshadrach added Transformations e.g. cumsum, diff, rank Enhancement labels Jun 9, 2026

rhshadrach added this to the 3.1 milestone Jun 9, 2026

hamdanal pushed a commit to hamdanal/pandas that referenced this pull request Jun 27, 2026

ENH: DataFrameGroupBy.transform now accepts list, dict, and NamedAgg …

e3acfbd

…arguments (GH#58318) (pandas-dev#65164)

		``output_name=NamedFunc(column, func)`` for named transformation or
		``output_name=NamedAgg(column, aggfunc)`` for named aggregation.

	``output_name=NamedFunc(column, func)`` for named transformation or
	``output_name=NamedAgg(column, aggfunc)`` for named aggregation.
	``output_name=NamedFunc(column, func)``.

	# ── dict / NamedAgg path ─────────────────────────────────────────
	# Also includes NamedAgg / NamedFunc

	# ── list path ────────────────────────────────────────────────────────
	# list path

Uh oh!

Uh oh!

Conversation

berasaikat commented Apr 11, 2026

Uh oh!

rhshadrach commented Apr 23, 2026

Uh oh!

berasaikat commented May 4, 2026

Uh oh!

rhshadrach left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

berasaikat commented May 10, 2026

Uh oh!

jbrockmendel commented May 11, 2026

Uh oh!

rhshadrach left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

berasaikat commented May 20, 2026

Uh oh!

rhshadrach left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rhshadrach May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rhshadrach left a comment

Choose a reason for hiding this comment

Uh oh!

jbrockmendel commented Jun 8, 2026

Uh oh!

Uh oh!

rhshadrach commented Jun 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

rhshadrach May 28, 2026 •

edited

Loading