Backport PR #35723: agg with list of non-aggregating functions (#35738)

Co-authored-by: Richard Shadrach <[email protected]>

Author: meeseeksmachine

doc/source/whatsnew/v1.1.1.rst

@@ -26,6 +26,7 @@ Fixed regressions
 - Fixed regression in :meth:`DataFrame.reset_index` would raise a ``ValueError`` on empty :class:`DataFrame` with a :class:`MultiIndex` with a ``datetime64`` dtype level (:issue:`35606`, :issue:`35657`)
 - Fixed regression where :meth:`DataFrame.merge_asof` would raise a ``UnboundLocalError`` when ``left_index`` , ``right_index`` and ``tolerance`` were set (:issue:`35558`)
 - Fixed regression in ``.groupby(..).rolling(..)`` where a custom ``BaseIndexer`` would be ignored (:issue:`35557`)
+- Fixed regression in :meth:`~pandas.core.groupby.DataFrameGroupBy.agg` where a list of functions would produce the wrong results if at least one of the functions did not aggregate. (:issue:`35490`)
 
 .. ---------------------------------------------------------------------------
 

pandas/core/groupby/generic.py

@@ -322,11 +322,14 @@ def _aggregate_multiple_funcs(self, arg):
             # let higher level handle
             return results
 
-        output = self._wrap_aggregated_output(results)
+        output = self._wrap_aggregated_output(results, index=None)
         return self.obj._constructor_expanddim(output, columns=columns)
 
+    # TODO: index should not be Optional - see GH 35490
     def _wrap_series_output(
-        self, output: Mapping[base.OutputKey, Union[Series, np.ndarray]], index: Index,
+        self,
+        output: Mapping[base.OutputKey, Union[Series, np.ndarray]],
+        index: Optional[Index],
     ) -> Union[Series, DataFrame]:
         """
         Wraps the output of a SeriesGroupBy operation into the expected result.
@@ -335,7 +338,7 @@ def _wrap_series_output(
         ----------
         output : Mapping[base.OutputKey, Union[Series, np.ndarray]]
             Data to wrap.
-        index : pd.Index
+        index : pd.Index or None
             Index to apply to the output.
 
         Returns
@@ -363,8 +366,11 @@ def _wrap_series_output(
 
         return result
 
+    # TODO: Remove index argument, use self.grouper.result_index, see GH 35490
     def _wrap_aggregated_output(
-        self, output: Mapping[base.OutputKey, Union[Series, np.ndarray]]
+        self,
+        output: Mapping[base.OutputKey, Union[Series, np.ndarray]],
+        index: Optional[Index],
     ) -> Union[Series, DataFrame]:
         """
         Wraps the output of a SeriesGroupBy aggregation into the expected result.
@@ -383,9 +389,7 @@ def _wrap_aggregated_output(
         In the vast majority of cases output will only contain one element.
         The exception is operations that expand dimensions, like ohlc.
         """
-        result = self._wrap_series_output(
-            output=output, index=self.grouper.result_index
-        )
+        result = self._wrap_series_output(output=output, index=index)
         return self._reindex_output(result)
 
     def _wrap_transformed_output(
@@ -1714,7 +1718,9 @@ def _insert_inaxis_grouper_inplace(self, result):
                 result.insert(0, name, lev)
 
     def _wrap_aggregated_output(
-        self, output: Mapping[base.OutputKey, Union[Series, np.ndarray]]
+        self,
+        output: Mapping[base.OutputKey, Union[Series, np.ndarray]],
+        index: Optional[Index],
     ) -> DataFrame:
         """
         Wraps the output of DataFrameGroupBy aggregations into the expected result.
@@ -1739,8 +1745,7 @@ def _wrap_aggregated_output(
             self._insert_inaxis_grouper_inplace(result)
             result = result._consolidate()
         else:
-            index = self.grouper.result_index
-            result.index = index
+            result.index = self.grouper.result_index
 
         if self.axis == 1:
             result = result.T

pandas/core/groupby/groupby.py

@@ -973,7 +973,9 @@ def _cython_transform(self, how: str, numeric_only: bool = True, **kwargs):
 
         return self._wrap_transformed_output(output)
 
-    def _wrap_aggregated_output(self, output: Mapping[base.OutputKey, np.ndarray]):
+    def _wrap_aggregated_output(
+        self, output: Mapping[base.OutputKey, np.ndarray], index: Optional[Index]
+    ):
         raise AbstractMethodError(self)
 
     def _wrap_transformed_output(self, output: Mapping[base.OutputKey, np.ndarray]):
@@ -1048,7 +1050,7 @@ def _cython_agg_general(
         if len(output) == 0:
             raise DataError("No numeric types to aggregate")
 
-        return self._wrap_aggregated_output(output)
+        return self._wrap_aggregated_output(output, index=self.grouper.result_index)
 
     def _python_agg_general(
         self, func, *args, engine="cython", engine_kwargs=None, **kwargs
@@ -1101,7 +1103,7 @@ def _python_agg_general(
 
                 output[key] = maybe_cast_result(values[mask], result)
 
-        return self._wrap_aggregated_output(output)
+        return self._wrap_aggregated_output(output, index=self.grouper.result_index)
 
     def _concat_objects(self, keys, values, not_indexed_same: bool = False):
         from pandas.core.reshape.concat import concat
@@ -2521,7 +2523,7 @@ def _get_cythonized_result(
             raise TypeError(error_msg)
 
         if aggregate:
-            return self._wrap_aggregated_output(output)
+            return self._wrap_aggregated_output(output, index=self.grouper.result_index)
         else:
             return self._wrap_transformed_output(output)
 

pandas/tests/groupby/aggregate/test_aggregate.py

@@ -1061,3 +1061,16 @@ def test_groupby_get_by_index():
     res = df.groupby("A").agg({"B": lambda x: x.get(x.index[-1])})
     expected = pd.DataFrame(dict(A=["S", "W"], B=[1.0, 2.0])).set_index("A")
     pd.testing.assert_frame_equal(res, expected)
+
+
+def test_nonagg_agg():
+    # GH 35490 - Single/Multiple agg of non-agg function give same results
+    # TODO: agg should raise for functions that don't aggregate
+    df = pd.DataFrame({"a": [1, 1, 2, 2], "b": [1, 2, 2, 1]})
+    g = df.groupby("a")
+
+    result = g.agg(["cumsum"])
+    result.columns = result.columns.droplevel(-1)
+    expected = g.agg("cumsum")
+
+    tm.assert_frame_equal(result, expected)