BUG: x in MultiIndex.drop(x) (#19054)

Author: databasedav

doc/source/whatsnew/v0.23.0.txt

@@ -398,6 +398,7 @@ Indexing
 - Bug in indexing non-scalar value from ``Series`` having non-unique ``Index`` will return value flattened (:issue:`17610`)
 - Bug in :func:`DatetimeIndex.insert` where inserting ``NaT`` into a timezone-aware index incorrectly raised (:issue:`16357`)
 - Bug in ``__setitem__`` when indexing a :class:`DataFrame` with a 2-d boolean ndarray (:issue:`18582`)
+- Bug in :func:`MultiIndex.__contains__` where non-tuple keys would return ``True`` even if they had been dropped (:issue:`19027`)
 - Bug in :func:`MultiIndex.set_labels` which would cause casting (and potentially clipping) of the new labels if the ``level`` argument is not 0 or a list like [0, 1, ... ]  (:issue:`19057`)
 - Bug in ``str.extractall`` when there were no matches empty :class:`Index` was returned instead of appropriate :class:`MultiIndex` (:issue:`19034`)
 

pandas/core/indexes/multi.py

@@ -2123,6 +2123,11 @@ def _maybe_to_slice(loc):
 
         if not isinstance(key, tuple):
             loc = self._get_level_indexer(key, level=0)
+
+            # _get_level_indexer returns an empty slice if the key has
+            # been dropped from the MultiIndex
+            if isinstance(loc, slice) and loc.start == loc.stop:
+                raise KeyError(key)
             return _maybe_to_slice(loc)
 
         keylen = len(key)

pandas/core/reshape/reshape.py

@@ -628,7 +628,11 @@ def _convert_level_number(level_num, columns):
     levsize = len(level_labels)
     drop_cols = []
     for key in unique_groups:
-        loc = this.columns.get_loc(key)
+        try:
+            loc = this.columns.get_loc(key)
+        except KeyError:
+            drop_cols.append(key)
+            continue
 
         # can make more efficient?
         # we almost always return a slice
@@ -639,10 +643,7 @@ def _convert_level_number(level_num, columns):
         else:
             slice_len = loc.stop - loc.start
 
-        if slice_len == 0:
-            drop_cols.append(key)
-            continue
-        elif slice_len != levsize:
+        if slice_len != levsize:
             chunk = this.loc[:, this.columns[loc]]
             chunk.columns = level_vals.take(chunk.columns.labels[-1])
             value_slice = chunk.reindex(columns=level_vals_used).values

pandas/tests/frame/test_mutate_columns.py

@@ -193,9 +193,10 @@ def test_delitem_multiindex(self):
         with pytest.raises(KeyError):
             del df[('A',)]
 
-        # xref: https://github.com./pandas-dev/pandas/issues/2770
-        # the 'A' is STILL in the columns!
-        assert 'A' in df.columns
+        # behavior of dropped/deleted MultiIndex levels changed from
+        # GH 2770 to GH 19027: MultiIndex no longer '.__contains__'
+        # levels which are dropped/deleted
+        assert 'A' not in df.columns
         with pytest.raises(KeyError):
             del df['A']
 

pandas/tests/indexing/test_multiindex.py

@@ -705,6 +705,26 @@ def test_multiindex_symmetric_difference(self):
         result = idx ^ idx2
         assert result.names == [None, None]
 
+    def test_multiindex_contains_dropped(self):
+        # GH 19027
+        # test that dropped MultiIndex levels are not in the MultiIndex
+        # despite continuing to be in the MultiIndex's levels
+        idx = MultiIndex.from_product([[1, 2], [3, 4]])
+        assert 2 in idx
+        idx = idx.drop(2)
+
+        # drop implementation keeps 2 in the levels
+        assert 2 in idx.levels[0]
+        # but it should no longer be in the index itself
+        assert 2 not in idx
+
+        # also applies to strings
+        idx = MultiIndex.from_product([['a', 'b'], ['c', 'd']])
+        assert 'a' in idx
+        idx = idx.drop('a')
+        assert 'a' in idx.levels[0]
+        assert 'a' not in idx
+
 
 class TestMultiIndexSlicers(object):