REF: avoid special-casing inside DTA/TDA.mean (pandas-dev#37422)

Author: jbrockmendel

pandas/core/arrays/datetimelike.py

@@ -1304,7 +1304,7 @@ def max(self, axis=None, skipna=True, *args, **kwargs):
         # Don't have to worry about NA `result`, since no NA went in.
         return self._box_func(result)
 
-    def mean(self, skipna=True):
+    def mean(self, skipna=True, axis: Optional[int] = 0):
         """
         Return the mean value of the Array.
 
@@ -1314,6 +1314,7 @@ def mean(self, skipna=True):
         ----------
         skipna : bool, default True
             Whether to ignore any NaT elements.
+        axis : int, optional, default 0
 
         Returns
         -------
@@ -1337,21 +1338,12 @@ def mean(self, skipna=True):
                 "obj.to_timestamp(how='start').mean()"
             )
 
-        mask = self.isna()
-        if skipna:
-            values = self[~mask]
-        elif mask.any():
-            return NaT
-        else:
-            values = self
-
-        if not len(values):
-            # short-circuit for empty max / min
-            return NaT
-
-        result = nanops.nanmean(values.view("i8"), skipna=skipna)
-        # Don't have to worry about NA `result`, since no NA went in.
-        return self._box_func(result)
+        result = nanops.nanmean(
+            self._ndarray, axis=axis, skipna=skipna, mask=self.isna()
+        )
+        if axis is None or self.ndim == 1:
+            return self._box_func(result)
+        return self._from_backing_data(result)
 
     def median(self, axis: Optional[int] = None, skipna: bool = True, *args, **kwargs):
         nv.validate_median(args, kwargs)

pandas/core/nanops.py

@@ -339,7 +339,12 @@ def _wrap_results(result, dtype: DtypeObj, fill_value=None):
             assert not isna(fill_value), "Expected non-null fill_value"
             if result == fill_value:
                 result = np.nan
-            result = Timestamp(result, tz=tz)
+            if tz is not None:
+                result = Timestamp(result, tz=tz)
+            elif isna(result):
+                result = np.datetime64("NaT", "ns")
+            else:
+                result = np.int64(result).view("datetime64[ns]")
         else:
             # If we have float dtype, taking a view will give the wrong result
             result = result.astype(dtype)
@@ -386,8 +391,9 @@ def _na_for_min_count(
 
     if values.ndim == 1:
         return fill_value
+    elif axis is None:
+        return fill_value
     else:
-        assert axis is not None  # assertion to make mypy happy
         result_shape = values.shape[:axis] + values.shape[axis + 1 :]
 
         result = np.full(result_shape, fill_value, dtype=values.dtype)

pandas/tests/arrays/test_datetimes.py

@@ -9,6 +9,7 @@
 from pandas.core.dtypes.dtypes import DatetimeTZDtype
 
 import pandas as pd
+from pandas import NaT
 import pandas._testing as tm
 from pandas.core.arrays import DatetimeArray
 from pandas.core.arrays.datetimes import sequence_to_dt64ns
@@ -566,3 +567,54 @@ def test_median_2d(self, arr1d):
         result = arr.median(axis=1, skipna=False)
         expected = type(arr)._from_sequence([pd.NaT], dtype=arr.dtype)
         tm.assert_equal(result, expected)
+
+    def test_mean(self, arr1d):
+        arr = arr1d
+
+        # manually verified result
+        expected = arr[0] + 0.4 * pd.Timedelta(days=1)
+
+        result = arr.mean()
+        assert result == expected
+        result = arr.mean(skipna=False)
+        assert result is pd.NaT
+
+        result = arr.dropna().mean(skipna=False)
+        assert result == expected
+
+        result = arr.mean(axis=0)
+        assert result == expected
+
+    def test_mean_2d(self):
+        dti = pd.date_range("2016-01-01", periods=6, tz="US/Pacific")
+        dta = dti._data.reshape(3, 2)
+
+        result = dta.mean(axis=0)
+        expected = dta[1]
+        tm.assert_datetime_array_equal(result, expected)
+
+        result = dta.mean(axis=1)
+        expected = dta[:, 0] + pd.Timedelta(hours=12)
+        tm.assert_datetime_array_equal(result, expected)
+
+        result = dta.mean(axis=None)
+        expected = dti.mean()
+        assert result == expected
+
+    @pytest.mark.parametrize("skipna", [True, False])
+    def test_mean_empty(self, arr1d, skipna):
+        arr = arr1d[:0]
+
+        assert arr.mean(skipna=skipna) is NaT
+
+        arr2d = arr.reshape(0, 3)
+        result = arr2d.mean(axis=0, skipna=skipna)
+        expected = DatetimeArray._from_sequence([NaT, NaT, NaT], dtype=arr.dtype)
+        tm.assert_datetime_array_equal(result, expected)
+
+        result = arr2d.mean(axis=1, skipna=skipna)
+        expected = arr  # i.e. 1D, empty
+        tm.assert_datetime_array_equal(result, expected)
+
+        result = arr2d.mean(axis=None, skipna=skipna)
+        assert result is NaT

pandas/tests/arrays/test_timedeltas.py

@@ -177,7 +177,7 @@ def test_neg_freq(self):
 
 
 class TestReductions:
-    @pytest.mark.parametrize("name", ["std", "min", "max", "median"])
+    @pytest.mark.parametrize("name", ["std", "min", "max", "median", "mean"])
     @pytest.mark.parametrize("skipna", [True, False])
     def test_reductions_empty(self, name, skipna):
         tdi = pd.TimedeltaIndex([])
@@ -334,3 +334,37 @@ def test_median(self):
 
         result = tdi.median(skipna=False)
         assert result is pd.NaT
+
+    def test_mean(self):
+        tdi = pd.TimedeltaIndex(["0H", "3H", "NaT", "5H06m", "0H", "2H"])
+        arr = tdi._data
+
+        # manually verified result
+        expected = pd.Timedelta(arr.dropna()._ndarray.mean())
+
+        result = arr.mean()
+        assert result == expected
+        result = arr.mean(skipna=False)
+        assert result is pd.NaT
+
+        result = arr.dropna().mean(skipna=False)
+        assert result == expected
+
+        result = arr.mean(axis=0)
+        assert result == expected
+
+    def test_mean_2d(self):
+        tdi = pd.timedelta_range("14 days", periods=6)
+        tda = tdi._data.reshape(3, 2)
+
+        result = tda.mean(axis=0)
+        expected = tda[1]
+        tm.assert_timedelta_array_equal(result, expected)
+
+        result = tda.mean(axis=1)
+        expected = tda[:, 0] + pd.Timedelta(hours=12)
+        tm.assert_timedelta_array_equal(result, expected)
+
+        result = tda.mean(axis=None)
+        expected = tdi.mean()
+        assert result == expected