BUG Fixing columns dropped from multi index in group by transform GH4…

doc/source/whatsnew/v1.5.0.rst

@@ -1007,6 +1007,7 @@ Groupby/resample/rolling
 - Bug in :meth:`DataFrame.resample` reduction methods when used with ``on`` would attempt to aggregate the provided column (:issue:`47079`)
 - Bug in :meth:`DataFrame.groupby` and :meth:`Series.groupby` would not respect ``dropna=False`` when the input DataFrame/Series had a NaN values in a :class:`MultiIndex` (:issue:`46783`)
 - Bug in :meth:`DataFrameGroupBy.resample` raises ``KeyError`` when getting the result from a key list which misses the resample key (:issue:`47362`)
+- Bug in :meth:`DataFrame.groupby` would lose index columns when the dataframe is empty for transforms, like fillna (:issue:`47787`)
 -
 
 Reshaping

pandas/core/groupby/groupby.py

@@ -1020,6 +1020,11 @@ def curried(x):
                 return self.apply(curried)
 
             is_transform = name in base.transformation_kernels
+
+            # Transform needs to keep the same schema, including when empty
+            if is_transform and self._obj_with_exclusions.empty:
+                return self._obj_with_exclusions
+
             result = self._python_apply_general(
                 curried, self._obj_with_exclusions, is_transform=is_transform
             )

pandas/tests/groupby/test_groupby.py

@@ -2348,6 +2348,20 @@ def test_groupby_duplicate_index():
     tm.assert_series_equal(result, expected)
 
 
+def test_group_on_empty_multiindex(transformation_func):
+    # GH 47787
+    # With one row, those are transforms so the schema should be the same
+    df = DataFrame(data=[[1, 2, 3, 4]], columns=["col_1", "col_2", "col_3", "col_4"])
+    df = df.set_index(["col_1", "col_2"])
+    result = df.groupby(["col_1"]).fillna("")
+    assert df.index.names == result.index.names
+    # When empty, expect the same schema as well
+    df = DataFrame(data=[], columns=["col_1", "col_2", "col_3", "col_4"])
+    df = df.set_index(["col_1", "col_2"])
+    result = df.groupby(["col_1"]).fillna("")
+    assert df.index.names == result.index.names
+
+
 @pytest.mark.parametrize(
     "idx",
     [