REGR: MultiIndex.dtypes has regular Index instead of MultiIndex index (#47349)

Author: phofl

doc/source/whatsnew/v1.4.3.rst

@@ -15,6 +15,7 @@ including other versions of pandas.
 Fixed regressions
 ~~~~~~~~~~~~~~~~~
 - Fixed regression in :meth:`DataFrame.replace` when the replacement value was explicitly ``None`` when passed in a dictionary to ``to_replace`` also casting other columns to object dtype even when there were no values to replace (:issue:`46634`)
+- Fixed regression in representation of ``dtypes`` attribute of :class:`MultiIndex` (:issue:`46900`)
 - Fixed regression when setting values with :meth:`DataFrame.loc` updating :class:`RangeIndex` when index was set as new column and column was updated afterwards (:issue:`47128`)
 - Fixed regression in :meth:`DataFrame.nsmallest` led to wrong results when ``np.nan`` in the sorting column (:issue:`46589`)
 - Fixed regression in :func:`read_fwf` raising ``ValueError`` when ``widths`` was specified with ``usecols`` (:issue:`46580`)

pandas/core/indexes/multi.py

@@ -763,7 +763,7 @@ def dtypes(self) -> Series:
         from pandas import Series
 
         names = com.fill_missing_names([level.name for level in self.levels])
-        return Series([level.dtype for level in self.levels], index=names)
+        return Series([level.dtype for level in self.levels], index=Index(names))
 
     def __len__(self) -> int:
         return len(self.codes[0])

pandas/tests/indexes/multi/test_constructors.py

@@ -827,3 +827,13 @@ def test_multiindex_inference_consistency():
     mi = MultiIndex.from_tuples([(x,) for x in arr])
     lev = mi.levels[0]
     assert lev.dtype == object
+
+
+def test_dtype_representation():
+    # GH#46900
+    pmidx = MultiIndex.from_arrays([[1], ["a"]], names=[("a", "b"), ("c", "d")])
+    result = pmidx.dtypes
+    expected = Series(
+        ["int64", "object"], index=MultiIndex.from_tuples([("a", "b"), ("c", "d")])
+    )
+    tm.assert_series_equal(result, expected)