API: .astype td64->td64 give requested dtype (pandas-dev#48963)

* API: .astype td64->td64 give requested dtype

* fix missing import

* Update doc/source/whatsnew/v1.6.0.rst

Co-authored-by: Matthew Roeschke <[email protected]>

Co-authored-by: Matthew Roeschke <[email protected]>

Author: jbrockmendel

doc/source/whatsnew/v1.6.0.rst

@@ -119,6 +119,7 @@ Other API changes
 - Passing ``nanoseconds`` greater than 999 or less than 0 in :class:`Timestamp` now raises a ``ValueError`` (:issue:`48538`, :issue:`48255`)
 - :func:`read_csv`: specifying an incorrect number of columns with ``index_col`` of now raises ``ParserError`` instead of ``IndexError`` when using the c parser.
 - :meth:`DataFrame.astype`, :meth:`Series.astype`, and :meth:`DatetimeIndex.astype` casting datetime64 data to any of "datetime64[s]", "datetime64[ms]", "datetime64[us]" will return an object with the given resolution instead of coercing back to "datetime64[ns]" (:issue:`48928`)
+- :meth:`DataFrame.astype`, :meth:`Series.astype`, and :meth:`DatetimeIndex.astype` casting timedelta64 data to any of "timedelta64[s]", "timedelta64[ms]", "timedelta64[us]" will return an object with the given resolution instead of coercing to "float64" dtype (:issue:`48963`)
 -
 
 .. ---------------------------------------------------------------------------

pandas/core/arrays/timedeltas.py

@@ -20,7 +20,9 @@
     Tick,
     Timedelta,
     astype_overflowsafe,
+    get_unit_from_dtype,
     iNaT,
+    is_supported_unit,
     periods_per_second,
     to_offset,
 )
@@ -308,6 +310,18 @@ def astype(self, dtype, copy: bool = True):
         dtype = pandas_dtype(dtype)
 
         if dtype.kind == "m":
+            if dtype == self.dtype:
+                if copy:
+                    return self.copy()
+                return self
+
+            if is_supported_unit(get_unit_from_dtype(dtype)):
+                # unit conversion e.g. timedelta64[s]
+                res_values = astype_overflowsafe(self._ndarray, dtype, copy=False)
+                return type(self)._simple_new(
+                    res_values, dtype=res_values.dtype, freq=self.freq
+                )
+
             return astype_td64_unit_conversion(self._ndarray, dtype, copy=copy)
 
         return dtl.DatetimeLikeArrayMixin.astype(self, dtype, copy=copy)

pandas/core/dtypes/astype.py

@@ -136,6 +136,14 @@ def astype_nansafe(
             return arr.view(dtype)
 
         elif dtype.kind == "m":
+            # TODO(2.0): change to use the same logic as TDA.astype, i.e.
+            #  giving the requested dtype for supported units (s, ms, us, ns)
+            #  and doing the old convert-to-float behavior otherwise.
+            if is_supported_unit(get_unit_from_dtype(arr.dtype)):
+                from pandas.core.construction import ensure_wrapped_if_datetimelike
+
+                arr = ensure_wrapped_if_datetimelike(arr)
+                return arr.astype(dtype, copy=copy)
             return astype_td64_unit_conversion(arr, dtype, copy=copy)
 
         raise TypeError(f"cannot astype a timedelta from [{arr.dtype}] to [{dtype}]")

pandas/tests/frame/methods/test_astype.py

@@ -480,12 +480,19 @@ def test_astype_to_timedelta_unit_ns(self, unit):
     @pytest.mark.parametrize("unit", ["us", "ms", "s", "h", "m", "D"])
     def test_astype_to_timedelta_unit(self, unit):
         # coerce to float
-        # GH#19223
+        # GH#19223 until 2.0 used to coerce to float
         dtype = f"m8[{unit}]"
         arr = np.array([[1, 2, 3]], dtype=dtype)
         df = DataFrame(arr)
         result = df.astype(dtype)
-        expected = DataFrame(df.values.astype(dtype).astype(float))
+
+        if unit in ["m", "h", "D"]:
+            # We don't support these, so we use the old logic to convert to float
+            expected = DataFrame(df.values.astype(dtype).astype(float))
+        else:
+            tda = pd.core.arrays.TimedeltaArray._simple_new(arr, dtype=arr.dtype)
+            expected = DataFrame(tda)
+            assert (expected.dtypes == dtype).all()
 
         tm.assert_frame_equal(result, expected)
 

pandas/tests/indexes/timedeltas/test_timedelta.py

@@ -13,6 +13,7 @@
     timedelta_range,
 )
 import pandas._testing as tm
+from pandas.core.arrays import TimedeltaArray
 from pandas.core.indexes.api import Int64Index
 from pandas.tests.indexes.datetimelike import DatetimeLike
 
@@ -101,19 +102,26 @@ def test_fields(self):
         assert rng.days.name == "name"
 
     def test_freq_conversion_always_floating(self):
-        # even if we have no NaTs, we get back float64; this matches TDA and Series
+        # pre-2.0 td64 astype converted to float64. now for supported units
+        #  (s, ms, us, ns) this converts to the requested dtype.
+        # This matches TDA and Series
         tdi = timedelta_range("1 Day", periods=30)
 
         res = tdi.astype("m8[s]")
-        expected = Index((tdi.view("i8") / 10**9).astype(np.float64))
+        exp_values = np.asarray(tdi).astype("m8[s]")
+        exp_tda = TimedeltaArray._simple_new(
+            exp_values, dtype=exp_values.dtype, freq=tdi.freq
+        )
+        expected = Index(exp_tda)
+        assert expected.dtype == "m8[s]"
         tm.assert_index_equal(res, expected)
 
         # check this matches Series and TimedeltaArray
         res = tdi._data.astype("m8[s]")
-        tm.assert_numpy_array_equal(res, expected._values)
+        tm.assert_equal(res, expected._values)
 
         res = tdi.to_series().astype("m8[s]")
-        tm.assert_numpy_array_equal(res._values, expected._values)
+        tm.assert_equal(res._values, expected._values._with_freq(None))
 
     def test_freq_conversion(self, index_or_series):
 
@@ -131,6 +139,8 @@ def test_freq_conversion(self, index_or_series):
         )
         tm.assert_equal(result, expected)
 
+        # We don't support "D" reso, so we use the pre-2.0 behavior
+        #  casting to float64
         result = td.astype("timedelta64[D]")
         expected = index_or_series([31, 31, 31, np.nan])
         tm.assert_equal(result, expected)
@@ -141,5 +151,9 @@ def test_freq_conversion(self, index_or_series):
         )
         tm.assert_equal(result, expected)
 
+        exp_values = np.asarray(td).astype("m8[s]")
+        exp_tda = TimedeltaArray._simple_new(exp_values, dtype=exp_values.dtype)
+        expected = index_or_series(exp_tda)
+        assert expected.dtype == "m8[s]"
         result = td.astype("timedelta64[s]")
         tm.assert_equal(result, expected)

pandas/tests/indexing/test_loc.py

@@ -649,14 +649,14 @@ def test_loc_setitem_consistency_slice_column_len(self):
             )
 
         with tm.assert_produces_warning(None, match=msg):
-            # timedelta64[s] -> float64, so this cannot be done inplace, so
+            # timedelta64[m] -> float64, so this cannot be done inplace, so
             #  no warning
             df.loc[:, ("Respondent", "Duration")] = df.loc[
                 :, ("Respondent", "Duration")
-            ].astype("timedelta64[s]")
+            ].astype("timedelta64[m]")
 
         expected = Series(
-            [1380, 720, 840, 2160.0], index=df.index, name=("Respondent", "Duration")
+            [23.0, 12.0, 14.0, 36.0], index=df.index, name=("Respondent", "Duration")
         )
         tm.assert_series_equal(df[("Respondent", "Duration")], expected)