WIP: keep column levels when using apply after grouping (#16231)

doc/source/whatsnew/v0.20.3.txt

@@ -65,6 +65,7 @@ Plotting
 Groupby/Resample/Rolling
 ^^^^^^^^^^^^^^^^^^^^^^^^
 
+- Bug in groupby logic causing MultiIndex column levels to be lost (:issue:`16231`)
 
 
 Sparse

pandas/core/base.py

@@ -658,6 +658,7 @@ def _aggregate_multiple_funcs(self, arg, _level, _axis):
 
         # degenerate case
         if obj.ndim == 1:
+            names = obj.index.names
             for a in arg:
                 try:
                     colg = self._gotitem(obj.name, ndim=1, subset=obj)
@@ -673,6 +674,7 @@ def _aggregate_multiple_funcs(self, arg, _level, _axis):
 
         # multiples
         else:
+            names = obj.columns.names
             for col in obj:
                 try:
                     colg = self._gotitem(col, ndim=1, subset=obj[col])
@@ -691,7 +693,7 @@ def _aggregate_multiple_funcs(self, arg, _level, _axis):
             raise ValueError("no results")
 
         try:
-            return concat(results, keys=keys, axis=1)
+            return concat(results, keys=keys, axis=1, names=names)
         except TypeError:
 
             # we are concatting non-NDFrame objects,

pandas/core/groupby.py

@@ -3481,9 +3481,9 @@ def aggregate(self, arg, *args, **kwargs):
                     assert not args and not kwargs
                     result = self._aggregate_multiple_funcs(
                         [arg], _level=_level, _axis=self.axis)
-                    result.columns = Index(
-                        result.columns.levels[0],
-                        name=self._selected_obj.columns.name)
+                    result.columns = result.columns.droplevel(-1)
+                    if result.columns.nlevels == 1:
+                        result.columns.name = self._selected_obj.columns.name
                 except:
                     result = self._aggregate_generic(arg, *args, **kwargs)
 

pandas/tests/groupby/test_groupby.py

@@ -3626,6 +3626,22 @@ def test_func(x):
         tm.assert_frame_equal(result1, expected1)
         tm.assert_frame_equal(result2, expected2)
 
+    @pytest.mark.parametrize('nlevel', range(1, 6))
+    @pytest.mark.parametrize('as_index', [False, True])
+    def test_groupby_aggregate_preserves_multiindex_columns(self, nlevel,
+                                                            as_index):
+        # GH 16231
+        cols = pd.MultiIndex.from_tuples([[i] * nlevel for i in range(2)],
+                                         names=['lev_{}'.format(lev)
+                                                for lev in range(nlevel)])
+        df = pd.DataFrame(np.random.randn(10, len(cols)), columns=cols)
+
+        grouped = df.groupby(df.index % 3, as_index=as_index)
+        via_direct = grouped.sum()
+        via_agg = grouped.aggregate(lambda x: x.sum())
+
+        tm.assert_frame_equal(via_direct, via_agg)
+
     def test_groupby_preserves_sort(self):
         # Test to ensure that groupby always preserves sort order of original
         # object. Issue #8588 and #9651

pandas/tests/test_resample.py

@@ -2972,6 +2972,37 @@ def f(x):
         result = g.apply(f)
         assert_frame_equal(result, expected)
 
+    def test_apply_preserves_multiindex_columns(self):
+        # GH 16231
+        # the original failing case
+        cols = pd.MultiIndex.from_tuples([('A', 'a', '', 'one'),
+                                          ('B', 'b', 'i', 'two')])
+        ind = pd.DatetimeIndex(start='2017-01-01', freq='15Min', periods=8)
+        df = pd.DataFrame(np.random.randn(8, 2), index=ind, columns=cols)
+
+        agg_dict = {col: (np.sum if col[3] == 'one' else np.mean)
+                    for col in df.columns}
+        resampled = df.resample('H').apply(lambda x: agg_dict[x.name](x))
+        assert isinstance(resampled.columns, pd.MultiIndex)
+
+    @pytest.mark.parametrize('nlevel', range(1, 6))
+    @pytest.mark.parametrize('ncol', [1, 2])
+    @pytest.mark.parametrize('freq', ['D', '360Min'])
+    def test_apply_preserves_multiindex_columns_grid(self, nlevel, ncol, freq):
+        # GH 16231
+        cols = pd.MultiIndex.from_tuples([[i] * nlevel for i in range(ncol)],
+                                         names=['lev_{}'.format(lev)
+                                                for lev in range(nlevel)])
+        idx = pd.date_range('2000-01-01', freq="H", periods=50)
+        df = pd.DataFrame(np.random.randn(len(idx), len(cols)),
+                          columns=cols, index=idx)
+
+        resampled = df.resample(freq)
+
+        via_direct = resampled.sum()
+        via_apply = resampled.apply(lambda x: x.sum())
+        tm.assert_frame_equal(via_direct, via_apply)
+
     def test_resample_groupby_with_label(self):
         # GH 13235
         index = date_range('2000-01-01', freq='2D', periods=5)