BUG: get_level_values() on int level upcasts to Float64Index if needed (#17930)

closes #17924

Author: toobaz

doc/source/whatsnew/v0.22.0.txt

@@ -189,7 +189,7 @@ Other API Changes
 - :func:`pandas.DataFrame.merge` no longer casts a ``float`` column to ``object`` when merging on ``int`` and ``float`` columns (:issue:`16572`)
 - The default NA value for :class:`UInt64Index` has changed from 0 to ``NaN``, which impacts methods that mask with NA, such as ``UInt64Index.where()`` (:issue:`18398`)
 - Refactored ``setup.py`` to use ``find_packages`` instead of explicitly listing out all subpackages (:issue:`18535`)
-- Rearranged the order of keyword arguments in :func:`read_excel()` to align with :func:`read_csv()` (:pr:`16672`)
+- Rearranged the order of keyword arguments in :func:`read_excel()` to align with :func:`read_csv()` (:issue:`16672`)
 - :func:`pandas.merge` now raises a ``ValueError`` when trying to merge on incompatible data types (:issue:`9780`)
 - :func:`wide_to_long` previously kept numeric-like suffixes as ``object`` dtype. Now they are cast to numeric if possible (:issue:`17627`)
 
@@ -266,6 +266,7 @@ Indexing
 
 - Bug in :func:`Series.truncate` which raises ``TypeError`` with a monotonic ``PeriodIndex`` (:issue:`17717`)
 - Bug in :func:`DataFrame.groupby` where tuples were interpreted as lists of keys rather than as keys (:issue:`17979`, :issue:`18249`)
+- Bug in :func:`MultiIndex.get_level_values` which would return an invalid index on level of ints with missing values (:issue:`17924`)
 - Bug in :func:`MultiIndex.remove_unused_levels` which would fill nan values (:issue:`18417`)
 - Bug in :func:`MultiIndex.from_tuples`` which would fail to take zipped tuples in python3 (:issue:`18434`)
 - Bug in :class:`Index` construction from list of mixed type tuples (:issue:`18505`)

pandas/core/indexes/numeric.py

@@ -62,6 +62,14 @@ def _maybe_cast_slice_bound(self, label, side, kind):
         # we will try to coerce to integers
         return self._maybe_cast_indexer(label)
 
+    @Appender(_index_shared_docs['_shallow_copy'])
+    def _shallow_copy(self, values=None, **kwargs):
+        if values is not None and not self._can_hold_na:
+            # Ensure we are not returning an Int64Index with float data:
+            return self._shallow_copy_with_infer(values=values, **kwargs)
+        return (super(NumericIndex, self)._shallow_copy(values=values,
+                                                        **kwargs))
+
     def _convert_for_op(self, value):
         """ Convert value to be insertable to ndarray """
 
@@ -395,9 +403,11 @@ def __contains__(self, other):
             try:
                 return len(other) <= 1 and ibase._try_get_item(other) in self
             except TypeError:
-                return False
-        except:
-            return False
+                pass
+        except TypeError:
+            pass
+
+        return False
 
     @Appender(_index_shared_docs['get_loc'])
     def get_loc(self, key, method=None, tolerance=None):

pandas/tests/indexes/test_multi.py

@@ -997,8 +997,8 @@ def test_get_level_values(self):
         exp = CategoricalIndex([1, 2, 3, 1, 2, 3])
         tm.assert_index_equal(index.get_level_values(1), exp)
 
-    @pytest.mark.xfail(reason='GH 17924 (returns Int64Index with float data)')
     def test_get_level_values_int_with_na(self):
+        # GH 17924
         arrays = [['a', 'b', 'b'], [1, np.nan, 2]]
         index = pd.MultiIndex.from_arrays(arrays)
         result = index.get_level_values(1)
@@ -1024,14 +1024,27 @@ def test_get_level_values_na(self):
 
         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_index_equal(values, expected)
+        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)
+
+    def test_get_level_values_all_na(self):
+        # GH 17924 when level entirely consists of nan
+        arrays = [[np.nan, np.nan, np.nan], ['a', np.nan, 1]]
+        index = pd.MultiIndex.from_arrays(arrays)
+        result = index.get_level_values(0)
+        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