BUG: dtype of DataFrame.idxmax/idxmin incorrect for empty frames

ci/code_checks.sh

@@ -172,16 +172,6 @@ if [[ -z "$CHECK" || "$CHECK" == "docstrings" ]]; then
         pandas.Period.asfreq \
         pandas.Period.now \
         pandas.arrays.PeriodArray \
-        pandas.Interval.closed \
-        pandas.Interval.left \
-        pandas.Interval.length \
-        pandas.Interval.right \
-        pandas.arrays.IntervalArray.left \
-        pandas.arrays.IntervalArray.right \
-        pandas.arrays.IntervalArray.closed \
-        pandas.arrays.IntervalArray.mid \
-        pandas.arrays.IntervalArray.length \
-        pandas.arrays.IntervalArray.is_non_overlapping_monotonic \
         pandas.arrays.IntervalArray.from_arrays \
         pandas.arrays.IntervalArray.to_tuples \
         pandas.Int8Dtype \
@@ -310,9 +300,7 @@ if [[ -z "$CHECK" || "$CHECK" == "docstrings" ]]; then
         pandas.CategoricalIndex.as_ordered \
         pandas.CategoricalIndex.as_unordered \
         pandas.CategoricalIndex.equals \
-        pandas.IntervalIndex.closed \
         pandas.IntervalIndex.values \
-        pandas.IntervalIndex.is_non_overlapping_monotonic \
         pandas.IntervalIndex.to_tuples \
         pandas.MultiIndex.dtypes \
         pandas.MultiIndex.drop \

doc/source/whatsnew/v2.1.0.rst

@@ -429,6 +429,7 @@ Reshaping
 ^^^^^^^^^
 - Bug in :func:`crosstab` when ``dropna=False`` would not keep ``np.nan`` in the result (:issue:`10772`)
 - Bug in :meth:`DataFrame.agg` and :meth:`Series.agg` on non-unique columns would return incorrect type when dist-like argument passed in (:issue:`51099`)
+- Bug in :meth:`DataFrame.idxmin` and :meth:`DataFrame.idxmax`, where the axis dtype would be lost for empty frames (:issue:`53265`)
 - Bug in :meth:`DataFrame.merge` not merging correctly when having ``MultiIndex`` with single level (:issue:`52331`)
 - Bug in :meth:`DataFrame.stack` losing extension dtypes when columns is a :class:`MultiIndex` and frame contains mixed dtypes (:issue:`45740`)
 - Bug in :meth:`DataFrame.transpose` inferring dtype for object column (:issue:`51546`)

pandas/_libs/interval.pyx

@@ -188,6 +188,14 @@ cdef class IntervalMixin:
         See Also
         --------
         Interval.is_empty : Indicates if an interval contains no points.
+
+        Examples
+        --------
+        >>> interval = pd.Interval(left=1, right=2, closed='left')
+        >>> interval
+        Interval(1, 2, closed='left')
+        >>> interval.length
+        1
         """
         return self.right - self.left
 
@@ -369,18 +377,42 @@ cdef class Interval(IntervalMixin):
     cdef readonly object left
     """
     Left bound for the interval.
+
+    Examples
+    --------
+    >>> interval = pd.Interval(left=1, right=2, closed='left')
+    >>> interval
+    Interval(1, 2, closed='left')
+    >>> interval.left
+    1
     """
 
     cdef readonly object right
     """
     Right bound for the interval.
+
+    Examples
+    --------
+    >>> interval = pd.Interval(left=1, right=2, closed='left')
+    >>> interval
+    Interval(1, 2, closed='left')
+    >>> interval.right
+    2
     """
 
     cdef readonly str closed
     """
     String describing the inclusive side the intervals.
 
     Either ``left``, ``right``, ``both`` or ``neither``.
+
+    Examples
+    --------
+    >>> interval = pd.Interval(left=1, right=2, closed='left')
+    >>> interval
+    Interval(1, 2, closed='left')
+    >>> interval.closed
+    'left'
     """
 
     def __init__(self, left, right, str closed="right"):

pandas/core/arrays/interval.py

@@ -1266,6 +1266,17 @@ def _format_space(self) -> str:
     def left(self):
         """
         Return the left endpoints of each Interval in the IntervalArray as an Index.
+
+        Examples
+        --------
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1), pd.Interval(2, 5)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0, 1], (2, 5]]
+        Length: 2, dtype: interval[int64, right]
+        >>> interv_arr.left
+        Index([0, 2], dtype='int64')
         """
         from pandas import Index
 
@@ -1275,6 +1286,17 @@ def left(self):
     def right(self):
         """
         Return the right endpoints of each Interval in the IntervalArray as an Index.
+
+        Examples
+        --------
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1), pd.Interval(2, 5)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0, 1], (2, 5]]
+        Length: 2, dtype: interval[int64, right]
+        >>> interv_arr.right
+        Index([1, 5], dtype='int64')
         """
         from pandas import Index
 
@@ -1284,13 +1306,35 @@ def right(self):
     def length(self) -> Index:
         """
         Return an Index with entries denoting the length of each Interval.
+
+        Examples
+        --------
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1), pd.Interval(1, 5)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0, 1], (1, 5]]
+        Length: 2, dtype: interval[int64, right]
+        >>> interv_arr.length
+        Index([1, 4], dtype='int64')
         """
         return self.right - self.left
 
     @property
     def mid(self) -> Index:
         """
         Return the midpoint of each Interval in the IntervalArray as an Index.
+
+        Examples
+        --------
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1), pd.Interval(1, 5)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0, 1], (1, 5]]
+        Length: 2, dtype: interval[int64, right]
+        >>> interv_arr.mid
+        Index([0.5, 3.0], dtype='float64')
         """
         try:
             return 0.5 * (self.left + self.right)
@@ -1378,6 +1422,27 @@ def closed(self) -> IntervalClosedType:
         String describing the inclusive side the intervals.
 
         Either ``left``, ``right``, ``both`` or ``neither``.
+
+        Examples
+        --------
+
+        For arrays:
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1), pd.Interval(1, 5)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0, 1], (1, 5]]
+        Length: 2, dtype: interval[int64, right]
+        >>> interv_arr.closed
+        'right'
+
+        For Interval Index:
+
+        >>> interv_idx = pd.interval_range(start=0, end=2)
+        >>> interv_idx
+        IntervalIndex([(0, 1], (1, 2]], dtype='interval[int64, right]')
+        >>> interv_idx.closed
+        'right'
         """
         return self.dtype.closed
 
@@ -1436,6 +1501,41 @@ def set_closed(self, closed: IntervalClosedType) -> Self:
 
         Non-overlapping means (no Intervals share points), and monotonic means
         either monotonic increasing or monotonic decreasing.
+
+        Examples
+        --------
+        For arrays:
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1), pd.Interval(1, 5)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0, 1], (1, 5]]
+        Length: 2, dtype: interval[int64, right]
+        >>> interv_arr.is_non_overlapping_monotonic
+        True
+
+        >>> interv_arr = pd.arrays.IntervalArray([pd.Interval(0, 1),
+        ...                                       pd.Interval(-1, 0.1)])
+        >>> interv_arr
+        <IntervalArray>
+        [(0.0, 1.0], (-1.0, 0.1]]
+        Length: 2, dtype: interval[float64, right]
+        >>> interv_arr.is_non_overlapping_monotonic
+        False
+
+        For Interval Index:
+
+        >>> interv_idx = pd.interval_range(start=0, end=2)
+        >>> interv_idx
+        IntervalIndex([(0, 1], (1, 2]], dtype='interval[int64, right]')
+        >>> interv_idx.is_non_overlapping_monotonic
+        True
+
+        >>> interv_idx = pd.interval_range(start=0, end=2, closed='both')
+        >>> interv_idx
+        IntervalIndex([[0, 1], [1, 2]], dtype='interval[int64, both]')
+        >>> interv_idx.is_non_overlapping_monotonic
+        False
         """
 
     @property

pandas/core/frame.py

@@ -11162,6 +11162,11 @@ def idxmin(
         self, axis: Axis = 0, skipna: bool = True, numeric_only: bool = False
     ) -> Series:
         axis = self._get_axis_number(axis)
+
+        if self.empty and len(self.axes[axis]):
+            axis_dtype = self.axes[axis].dtype
+            return self._constructor_sliced(dtype=axis_dtype)
+
         if numeric_only:
             data = self._get_numeric_data()
         else:
@@ -11187,6 +11192,11 @@ def idxmax(
         self, axis: Axis = 0, skipna: bool = True, numeric_only: bool = False
     ) -> Series:
         axis = self._get_axis_number(axis)
+
+        if self.empty and len(self.axes[axis]):
+            axis_dtype = self.axes[axis].dtype
+            return self._constructor_sliced(dtype=axis_dtype)
+
         if numeric_only:
             data = self._get_numeric_data()
         else:

pandas/tests/frame/test_reductions.py

@@ -964,6 +964,18 @@ def test_idxmin(self, float_frame, int_frame, skipna, axis):
             expected = df.apply(Series.idxmin, axis=axis, skipna=skipna)
             tm.assert_series_equal(result, expected)
 
+    @pytest.mark.parametrize("axis", [0, 1])
+    def test_idxmin_empty(self, index, skipna, axis):
+        # GH53265
+        if axis == 0:
+            frame = DataFrame(index=index)
+        else:
+            frame = DataFrame(columns=index)
+
+        result = frame.idxmin(axis=axis, skipna=skipna)
+        expected = Series(dtype=index.dtype)
+        tm.assert_series_equal(result, expected)
+
     @pytest.mark.parametrize("numeric_only", [True, False])
     def test_idxmin_numeric_only(self, numeric_only):
         df = DataFrame({"a": [2, 3, 1], "b": [2, 1, 1], "c": list("xyx")})
@@ -992,6 +1004,18 @@ def test_idxmax(self, float_frame, int_frame, skipna, axis):
             expected = df.apply(Series.idxmax, axis=axis, skipna=skipna)
             tm.assert_series_equal(result, expected)
 
+    @pytest.mark.parametrize("axis", [0, 1])
+    def test_idxmax_empty(self, index, skipna, axis):
+        # GH53265
+        if axis == 0:
+            frame = DataFrame(index=index)
+        else:
+            frame = DataFrame(columns=index)
+
+        result = frame.idxmax(axis=axis, skipna=skipna)
+        expected = Series(dtype=index.dtype)
+        tm.assert_series_equal(result, expected)
+
     @pytest.mark.parametrize("numeric_only", [True, False])
     def test_idxmax_numeric_only(self, numeric_only):
         df = DataFrame({"a": [2, 3, 1], "b": [2, 1, 1], "c": list("xyx")})