BUG: fix dtype of all-NaN MultiIndex level

closes #17929

Author: toobaz

doc/source/whatsnew/v0.21.0.txt

@@ -964,6 +964,7 @@ Indexing
 - When called on an unsorted ``MultiIndex``, the ``loc`` indexer now will raise ``UnsortedIndexError`` only if proper slicing is used on non-sorted levels (:issue:`16734`).
 - Fixes regression in 0.20.3 when indexing with a string on a ``TimedeltaIndex`` (:issue:`16896`).
 - Fixed :func:`TimedeltaIndex.get_loc` handling of ``np.timedelta64`` inputs (:issue:`16909`).
+- Bug in ``MultiIndex`` which would assign object dtype to all-NaN levels (:issue:`17929`).
 - Fix :func:`MultiIndex.sort_index` ordering when ``ascending`` argument is a list, but not all levels are specified, or are in a different order (:issue:`16934`).
 - Fixes bug where indexing with ``np.inf`` caused an ``OverflowError`` to be raised (:issue:`16957`)
 - Bug in reindexing on an empty ``CategoricalIndex`` (:issue:`16770`)

pandas/core/categorical.py

@@ -2276,6 +2276,7 @@ def _factorize_from_iterable(values):
         a CategoricalIndex keeping the categories and order of `values`.
     """
     from pandas.core.indexes.category import CategoricalIndex
+    from pandas.core.indexes.numeric import Float64Index
 
     if not is_list_like(values):
         raise TypeError("Input must be list-like")
@@ -2291,6 +2292,8 @@ def _factorize_from_iterable(values):
         cat = Categorical(values, ordered=True)
         categories = cat.categories
         codes = cat.codes
+        if len(codes) and not len(categories):
+            categories = Float64Index([])
     return codes, categories
 
 

pandas/tests/indexes/test_multi.py

@@ -970,12 +970,13 @@ def test_get_level_values_na(self):
 
         arrays = [[np.nan, np.nan, np.nan], ['a', np.nan, 1]]
         index = pd.MultiIndex.from_arrays(arrays)
-        values = index.get_level_values(0)
-        expected = np.array([np.nan, np.nan, np.nan])
-        tm.assert_numpy_array_equal(values.values.astype(float), expected)
-        values = index.get_level_values(1)
-        expected = np.array(['a', np.nan, 1], dtype=object)
-        tm.assert_numpy_array_equal(values.values, expected)
+        result = index.get_level_values(0)
+        expected = pd.Index([np.nan, np.nan, np.nan])
+        tm.assert_index_equal(result, expected)
+
+        result = index.get_level_values(1)
+        expected = pd.Index(['a', np.nan, 1], dtype=object)
+        tm.assert_index_equal(result, expected)
 
         arrays = [['a', 'b', 'b'], pd.DatetimeIndex([0, 1, pd.NaT])]
         index = pd.MultiIndex.from_arrays(arrays)