BUG: Fix BaseWindowGroupby.aggregate where as_index is ignored

doc/source/whatsnew/v2.2.0.rst

@@ -169,6 +169,7 @@ Bug fixes
 ~~~~~~~~~
 - Bug in :class:`AbstractHolidayCalendar` where timezone data was not propagated when computing holiday observances (:issue:`54580`)
 - Bug in :class:`pandas.core.window.Rolling` where duplicate datetimelike indexes are treated as consecutive rather than equal with ``closed='left'`` and ``closed='neither'`` (:issue:`20712`)
+- Bug in :meth:`DataFrameGroupBy.rolling.agg` where ``as_index`` is ignored with list-like and dictionary-like ``func`` parameters (:issue:`31007`)
 
 Categorical
 ^^^^^^^^^^^

pandas/core/window/ewm.py

@@ -887,7 +887,11 @@ def _cov(X, Y):
         )
 
 
-class ExponentialMovingWindowGroupby(BaseWindowGroupby, ExponentialMovingWindow):
+# error: Definition of "agg" in base class "BaseWindowGroupby" is
+# incompatible with definition in base class "ExponentialMovingWindow"
+class ExponentialMovingWindowGroupby(  # type: ignore[misc]
+    BaseWindowGroupby, ExponentialMovingWindow
+):
     """
     Provide an exponential moving window groupby implementation.
     """

pandas/core/window/expanding.py

@@ -942,7 +942,9 @@ def corr(
         )
 
 
-class ExpandingGroupby(BaseWindowGroupby, Expanding):
+# error: Definition of "agg" in base class "BaseWindowGroupby" is
+# incompatible with definition in base class "Expanding"
+class ExpandingGroupby(BaseWindowGroupby, Expanding):  # type: ignore[misc]
     """
     Provide a expanding groupby implementation.
     """

pandas/core/window/rolling.py

@@ -34,7 +34,9 @@
 from pandas.core.dtypes.common import (
     ensure_float64,
     is_bool,
+    is_dict_like,
     is_integer,
+    is_list_like,
     is_numeric_dtype,
     needs_i8_conversion,
 )
@@ -875,6 +877,14 @@ def _gotitem(self, key, ndim, subset=None):
             subset = self.obj.set_index(self._on)
         return super()._gotitem(key, ndim, subset=subset)
 
+    def aggregate(self, func, *args, **kwargs):
+        result = super().aggregate(func, *args, **kwargs)
+        if not self._as_index and (is_list_like(func) or is_dict_like(func)):
+            result = result.reset_index(level=list(range(len(self._grouper.names))))
+        return result
+
+    agg = aggregate
+
 
 class Window(BaseWindow):
     """
@@ -2852,7 +2862,9 @@ def corr(
 Rolling.__doc__ = Window.__doc__
 
 
-class RollingGroupby(BaseWindowGroupby, Rolling):
+# error: Definition of "agg" in base class "BaseWindowGroupby" is
+# incompatible with definition in base class "Rolling"
+class RollingGroupby(BaseWindowGroupby, Rolling):  # type: ignore[misc]
     """
     Provide a rolling groupby implementation.
     """

pandas/tests/window/test_groupby.py

@@ -894,6 +894,42 @@ def test_as_index_false(self, by, expected_data):
         )
         tm.assert_frame_equal(result, expected)
 
+    @pytest.mark.parametrize(
+        "f", ["mean", lambda x: x.mean(), {"agg_col": "mean"}, ["mean"]]
+    )
+    def test_aggregate_as_index_false(self, f):
+        # GH 31007
+        index = date_range(end="2020-01-01", periods=10)
+        groupby_col = ["A", "A", "A", "A", "A", "B", "B", "B", "B", "B"]
+        df = DataFrame(
+            {"groupby_col": groupby_col, "agg_col": [1, 1, 0, 1, 0, 0, 0, 0, 1, 0]},
+            index=index,
+        )
+
+        result = df.groupby("groupby_col", as_index=False).rolling(4).agg(f)
+        if isinstance(f, list):
+            result.columns = result.columns.get_level_values(0)
+
+        expected = DataFrame(
+            {
+                "groupby_col": groupby_col,
+                "agg_col": [
+                    np.nan,
+                    np.nan,
+                    np.nan,
+                    0.75,
+                    0.5,
+                    np.nan,
+                    np.nan,
+                    np.nan,
+                    0.25,
+                    0.25,
+                ],
+            },
+            index=index,
+        )
+        tm.assert_frame_equal(result, expected)
+
     def test_nan_and_zero_endpoints(self, any_int_numpy_dtype):
         # https://github.com/twosigma/pandas/issues/53
         typ = np.dtype(any_int_numpy_dtype).type