BUG: Bug in a groupby of a non-lexsorted MultiIndex and multiple grou…

…ping levels closes pandas-dev#14776
jreback · Nov 30, 2016 · cf31905 · cf31905
1 parent 2bd9c95
commit cf31905
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 2 deletions.
diff --git a/doc/source/whatsnew/v0.19.2.txt b/doc/source/whatsnew/v0.19.2.txt
@@ -36,7 +36,7 @@ Bug Fixes
 - Bug in ``pd.read_csv`` for the Python engine in which an unhelpful error message was being raised when ``skipfooter`` was not being respected by Python's CSV library (:issue:`13879`)
 
 
-
+- Bug in ``.groupby(..., sort=True)`` of a non-lexsorted MultiIndex when grouping with multiple levels (:issue:`14776`)
 
 
 

diff --git a/pandas/core/groupby.py b/pandas/core/groupby.py
@@ -861,7 +861,17 @@ def reset_identity(values):
             if isinstance(result, Series):
                 result = result.reindex(ax)
             else:
-                result = result.reindex_axis(ax, axis=self.axis)
+
+                # this is a very unfortunate situation
+                # we have a multi-index that is NOT lexsorted
+                # and we have a result which is duplicated
+                # we can't reindex, so we resort to this
+                # GH 14776
+                if isinstance(ax, MultiIndex) and not ax.is_unique:
+                    result = result.take(result.index.get_indexer_for(
+                        ax.values).unique(), axis=self.axis)
+                else:
+                    result = result.reindex_axis(ax, axis=self.axis)
 
         elif self.group_keys:
 

diff --git a/pandas/tests/test_groupby.py b/pandas/tests/test_groupby.py
@@ -4736,6 +4736,25 @@ def test_groupby_multiindex_not_lexsorted(self):
             result = not_lexsorted_df.groupby('a').mean()
         tm.assert_frame_equal(expected, result)
 
+        # a transforming function should work regardless of sort
+        # GH 14776
+        df = DataFrame({'x': ['a', 'a', 'b', 'a'],
+                        'y': [1, 1, 2, 2],
+                        'z': [1, 2, 3, 4]}).set_index(['x', 'y'])
+        self.assertFalse(df.index.is_lexsorted())
+
+        for level in [0, 1, [0, 1]]:
+            for sort in [False, True]:
+                result = df.groupby(level=level, sort=sort).apply(
+                    DataFrame.drop_duplicates)
+                expected = df
+                tm.assert_frame_equal(expected, result)
+
+                result = df.sort_index().groupby(level=level, sort=sort).apply(
+                    DataFrame.drop_duplicates)
+                expected = df.sort_index()
+                tm.assert_frame_equal(expected, result)
+
     def test_groupby_levels_and_columns(self):
         # GH9344, GH9049
         idx_names = ['x', 'y']