BUG: fix for "TypeError: unorderable types" when creating MultiIndex with mixed dtypes (#22072)

closes #15457

Author: dickreuter

doc/source/whatsnew/v0.24.0.txt

@@ -626,7 +626,7 @@ MultiIndex
 
 - Removed compatibility for :class:`MultiIndex` pickles prior to version 0.8.0; compatibility with :class:`MultiIndex` pickles from version 0.13 forward is maintained (:issue:`21654`)
 - :meth:`MultiIndex.get_loc_level` (and as a consequence, ``.loc`` on a :class:``MultiIndex``ed object) will now raise a ``KeyError``, rather than returning an empty ``slice``, if asked a label which is present in the ``levels`` but is unused (:issue:`22221`)
--
+- Fix ``TypeError`` in Python 3 when creating :class:`MultiIndex` in which some levels have mixed types, e.g. when some labels are tuples (:issue:`15457`)
 
 I/O
 ^^^

pandas/core/arrays/categorical.py

@@ -2538,7 +2538,10 @@ def _factorize_from_iterable(values):
                                       ordered=values.ordered)
         codes = values.codes
     else:
-        cat = Categorical(values, ordered=True)
+        # The value of ordered is irrelevant since we don't use cat as such,
+        # but only the resulting categories, the order of which is independent
+        # from ordered. Set ordered to False as default. See GH #15457
+        cat = Categorical(values, ordered=False)
         categories = cat.categories
         codes = cat.codes
     return codes, categories

pandas/tests/indexes/multi/test_constructor.py

@@ -463,3 +463,12 @@ def test_tuples_with_name_string():
         pd.Index(li, name='abc')
     with pytest.raises(ValueError):
         pd.Index(li, name='a')
+
+
+def test_from_tuples_with_tuple_label():
+    # GH 15457
+    expected = pd.DataFrame([[2, 1, 2], [4, (1, 2), 3]],
+                            columns=['a', 'b', 'c']).set_index(['a', 'b'])
+    idx = pd.MultiIndex.from_tuples([(2, 1), (4, (1, 2))], names=('a', 'b'))
+    result = pd.DataFrame([2, 3], columns=['c'], index=idx)
+    tm.assert_frame_equal(expected, result)