BUG: get_level_values() on int level upcasts to Float64Index if needed

closes pandas-dev#17924

Author: toobaz

doc/source/whatsnew/v0.21.0.txt

@@ -970,6 +970,7 @@ Indexing
 - Fixes ``DataFrame.loc`` for setting with alignment and tz-aware ``DatetimeIndex`` (:issue:`16889`)
 - Avoids ``IndexError`` when passing an Index or Series to ``.iloc`` with older numpy (:issue:`17193`)
 - Allow unicode empty strings as placeholders in multilevel columns in Python 2 (:issue:`17099`)
+- Bug in :meth:`MultiIndex.get_level_values` which would return an invalid index on level of ints with missing values (:issue:`17924`)
 - Bug in ``.iloc`` when used with inplace addition or assignment and an int indexer on a ``MultiIndex`` causing the wrong indexes to be read from and written to (:issue:`17148`)
 - Bug in ``.isin()`` in which checking membership in empty ``Series`` objects raised an error (:issue:`16991`)
 - Bug in ``CategoricalIndex`` reindexing in which specified indices containing duplicates were not being respected (:issue:`17323`)

pandas/core/indexes/multi.py

@@ -906,7 +906,12 @@ def _get_level_values(self, level):
         labels = self.labels[level]
         filled = algos.take_1d(unique._values, labels,
                                fill_value=unique._na_value)
-        values = unique._shallow_copy(filled)
+
+        if not unique._can_hold_na and filled.dtype == 'float':
+            values = unique._shallow_copy_with_infer(filled)
+        else:
+            values = unique._shallow_copy(filled)
+
         return values
 
     def get_level_values(self, level):

pandas/tests/indexes/test_multi.py

@@ -956,37 +956,43 @@ def test_get_level_values(self):
         tm.assert_index_equal(index.get_level_values(1), exp)
 
     def test_get_level_values_na(self):
+        # GH 17924
         arrays = [['a', 'b', 'b'], [1, np.nan, 2]]
         index = pd.MultiIndex.from_arrays(arrays)
-        values = index.get_level_values(1)
-        expected = np.array([1, np.nan, 2])
-        tm.assert_numpy_array_equal(values.values.astype(float), expected)
+        result = index.get_level_values(1)
+        expected = pd.Index([1, np.nan, 2])
+        tm.assert_index_equal(result, expected)
 
         arrays = [['a', 'b', 'b'], [np.nan, np.nan, 2]]
         index = pd.MultiIndex.from_arrays(arrays)
-        values = index.get_level_values(1)
-        expected = np.array([np.nan, np.nan, 2])
-        tm.assert_numpy_array_equal(values.values.astype(float), expected)
-
-        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(1)
+        expected = pd.Index([np.nan, np.nan, 2])
+        tm.assert_index_equal(result, expected)
 
         arrays = [['a', 'b', 'b'], pd.DatetimeIndex([0, 1, pd.NaT])]
         index = pd.MultiIndex.from_arrays(arrays)
-        values = index.get_level_values(1)
+        result = index.get_level_values(1)
         expected = pd.DatetimeIndex([0, 1, pd.NaT])
-        tm.assert_numpy_array_equal(values.values, expected.values)
+        tm.assert_index_equal(result, expected)
 
         arrays = [[], []]
         index = pd.MultiIndex.from_arrays(arrays)
-        values = index.get_level_values(0)
-        assert values.shape == (0, )
+        result = index.get_level_values(0)
+        expected = pd.Index([], dtype=object)
+        tm.assert_index_equal(result, expected)
+
+    @pytest.mark.xfail(reason='GH 17929 (empty level is casted to object')
+    def test_get_level_values_all_na(self):
+        arrays = [[np.nan, np.nan, np.nan], ['a', np.nan, 1]]
+        index = pd.MultiIndex.from_arrays(arrays)
+        result = index.get_level_values(0)
+        # workaround for #17929 (should be 'float')
+        expected = pd.Index([np.nan, np.nan, np.nan], dtype=np.float64)
+        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)
 
     def test_reorder_levels(self):
         # this blows up