Backport PR #53001 on branch 2.0.x (BUG: Series.describe treating pya…

…rrow timestamps/timedeltas as categorical) (#53031)

* Backport PR #53001: BUG: Series.describe treating pyarrow timestamps/timedeltas as categorical

* clean

doc/source/whatsnew/v2.0.2.rst

@@ -23,6 +23,7 @@ Bug fixes
 - Bug in :func:`api.interchange.from_dataframe` was returning :class:`DataFrame`'s of incorrect sizes when called on slices (:issue:`52824`)
 - Bug in :func:`api.interchange.from_dataframe` was unnecessarily raising on bitmasks (:issue:`49888`)
 - Bug in :meth:`DataFrame.convert_dtypes` ignores ``convert_*`` keywords when set to False ``dtype_backend="pyarrow"`` (:issue:`52872`)
+- Bug in :meth:`Series.describe` treating pyarrow-backed timestamps and timedeltas as categorical data (:issue:`53001`)
 - Bug in :meth:`pd.array` raising for ``NumPy`` array and ``pa.large_string`` or ``pa.large_binary`` (:issue:`52590`)
 -
 

pandas/core/methods/describe.py

@@ -31,11 +31,10 @@
 from pandas.core.dtypes.common import (
     is_bool_dtype,
     is_complex_dtype,
-    is_datetime64_any_dtype,
     is_extension_array_dtype,
     is_numeric_dtype,
-    is_timedelta64_dtype,
 )
+from pandas.core.dtypes.dtypes import DatetimeTZDtype
 
 from pandas.core.arrays.arrow.dtype import ArrowDtype
 from pandas.core.arrays.floating import Float64Dtype
@@ -232,9 +231,13 @@ def describe_numeric_1d(series: Series, percentiles: Sequence[float]) -> Series:
     dtype: DtypeObj | None
     if is_extension_array_dtype(series):
         if isinstance(series.dtype, ArrowDtype):
-            import pyarrow as pa
+            if series.dtype.kind == "m":
+                # GH53001: describe timedeltas with object dtype
+                dtype = None
+            else:
+                import pyarrow as pa
 
-            dtype = ArrowDtype(pa.float64())
+                dtype = ArrowDtype(pa.float64())
         else:
             dtype = Float64Dtype()
     elif is_numeric_dtype(series) and not is_complex_dtype(series):
@@ -362,9 +365,9 @@ def select_describe_func(
         return describe_categorical_1d
     elif is_numeric_dtype(data):
         return describe_numeric_1d
-    elif is_datetime64_any_dtype(data.dtype):
+    elif data.dtype.kind == "M" or isinstance(data.dtype, DatetimeTZDtype):
         return describe_timestamp_1d
-    elif is_timedelta64_dtype(data.dtype):
+    elif data.dtype.kind == "m":
         return describe_numeric_1d
     else:
         return describe_categorical_1d

pandas/tests/extension/test_arrow.py

@@ -2658,6 +2658,36 @@ def test_describe_numeric_data(pa_type):
     tm.assert_series_equal(result, expected)
 
 
+@pytest.mark.parametrize("pa_type", tm.TIMEDELTA_PYARROW_DTYPES)
+def test_describe_timedelta_data(pa_type):
+    # GH53001
+    data = pd.Series(range(1, 10), dtype=ArrowDtype(pa_type))
+    result = data.describe()
+    expected = pd.Series(
+        [9] + pd.to_timedelta([5, 2, 1, 3, 5, 7, 9], unit=pa_type.unit).tolist(),
+        dtype=object,
+        index=["count", "mean", "std", "min", "25%", "50%", "75%", "max"],
+    )
+    tm.assert_series_equal(result, expected)
+
+
+@pytest.mark.parametrize("pa_type", tm.DATETIME_PYARROW_DTYPES)
+def test_describe_datetime_data(pa_type):
+    # GH53001
+    data = pd.Series(range(1, 10), dtype=ArrowDtype(pa_type))
+    result = data.describe()
+    expected = pd.Series(
+        [9]
+        + [
+            pd.Timestamp(v, tz=pa_type.tz, unit=pa_type.unit)
+            for v in [5, 1, 3, 5, 7, 9]
+        ],
+        dtype=object,
+        index=["count", "mean", "min", "25%", "50%", "75%", "max"],
+    )
+    tm.assert_series_equal(result, expected)
+
+
 @pytest.mark.xfail(
     pa_version_under8p0,
     reason="Function 'add_checked' has no kernel matching input types",