DEPR: dtype inference in value_counts

doc/source/whatsnew/v2.2.0.rst

@@ -396,6 +396,7 @@ Other Deprecations
 - Deprecated the ``errors="ignore"`` option in :func:`to_datetime`, :func:`to_timedelta`, and :func:`to_numeric`; explicitly catch exceptions instead (:issue:`54467`)
 - Deprecated the ``fastpath`` keyword in the :class:`Series` constructor (:issue:`20110`)
 - Deprecated the ``ordinal`` keyword in :class:`PeriodIndex`, use :meth:`PeriodIndex.from_ordinals` instead (:issue:`55960`)
+- Deprecated the behavior of :meth:`Series.value_counts` and :meth:`Index.value_counts` with object dtype; in a future version these will not perform dtype inference on the resulting :class:`Index`, do ``result.index = result.index.infer_objects()`` to retain the old behavior (:issue:`56161`)
 - Deprecated the extension test classes ``BaseNoReduceTests``, ``BaseBooleanReduceTests``, and ``BaseNumericReduceTests``, use ``BaseReduceTests`` instead (:issue:`54663`)
 - Deprecated the option ``mode.data_manager`` and the ``ArrayManager``; only the ``BlockManager`` will be available in future versions (:issue:`55043`)
 - Deprecated the previous implementation of :class:`DataFrame.stack`; specify ``future_stack=True`` to adopt the future version (:issue:`53515`)

pandas/core/algorithms.py

@@ -932,6 +932,16 @@ def value_counts_internal(
             idx = Index(keys)
             if idx.dtype == bool and keys.dtype == object:
                 idx = idx.astype(object)
+            elif idx.dtype != keys.dtype:
+                warnings.warn(
+                    # GH#56161
+                    "The behavior of value_counts with object-dtype is deprecated. "
+                    "In a future version, this will *not* perform dtype inference "
+                    "on the resulting index. To retain the old behavior, use "
+                    "`result.index = result.index.infer_objects()`",
+                    FutureWarning,
+                    stacklevel=find_stack_level(),
+                )
             idx.name = index_name
 
             result = Series(counts, index=idx, name=name, copy=False)
@@ -1712,8 +1722,16 @@ def union_with_duplicates(
     """
     from pandas import Series
 
-    l_count = value_counts_internal(lvals, dropna=False)
-    r_count = value_counts_internal(rvals, dropna=False)
+    with warnings.catch_warnings():
+        # filter warning from object dtype inference; we will end up discarding
+        # the index here, so the deprecation does not affect the end result here.
+        warnings.filterwarnings(
+            "ignore",
+            "The behavior of value_counts with object-dtype is deprecated",
+            category=FutureWarning,
+        )
+        l_count = value_counts_internal(lvals, dropna=False)
+        r_count = value_counts_internal(rvals, dropna=False)
     l_count, r_count = l_count.align(r_count, fill_value=0)
     final_count = np.maximum(l_count.values, r_count.values)
     final_count = Series(final_count, index=l_count.index, dtype="int", copy=False)

pandas/core/arrays/interval.py

@@ -12,6 +12,7 @@
     Union,
     overload,
 )
+import warnings
 
 import numpy as np
 
@@ -1226,7 +1227,16 @@ def value_counts(self, dropna: bool = True) -> Series:
         Series.value_counts
         """
         # TODO: implement this is a non-naive way!
-        return value_counts(np.asarray(self), dropna=dropna)
+        with warnings.catch_warnings():
+            warnings.filterwarnings(
+                "ignore",
+                "The behavior of value_counts with object-dtype is deprecated",
+                category=FutureWarning,
+            )
+            result = value_counts(np.asarray(self), dropna=dropna)
+            # Once the deprecation is enforced, we will need to do
+            #  `result.index = result.index.astype(self.dtype)`
+        return result
 
     # ---------------------------------------------------------------------
     # Rendering Methods

pandas/tests/base/test_value_counts.py

@@ -336,3 +336,16 @@ def test_value_counts_with_nan(dropna, index_or_series):
     else:
         expected = Series([1, 1, 1], index=[True, pd.NA, np.nan], name="count")
     tm.assert_series_equal(res, expected)
+
+
+def test_value_counts_object_inference_deprecated():
+    # GH#56161
+    dti = pd.date_range("2016-01-01", periods=3, tz="UTC")
+
+    idx = dti.astype(object)
+    msg = "The behavior of value_counts with object-dtype is deprecated"
+    with tm.assert_produces_warning(FutureWarning, match=msg):
+        res = idx.value_counts()
+
+    exp = dti.value_counts()
+    tm.assert_series_equal(res, exp)