DEPR: special-casing dt64/td64 in Series.unique

Author: jbrockmendel

pandas/core/base.py

@@ -15,6 +15,7 @@
     cast,
     final,
 )
+import warnings
 
 import numpy as np
 
@@ -34,6 +35,7 @@
     cache_readonly,
     doc,
 )
+from pandas.util._exceptions import find_stack_level
 
 from pandas.core.dtypes.common import (
     is_categorical_dtype,
@@ -977,6 +979,14 @@ def unique(self):
             if self.dtype.kind in ["m", "M"] and isinstance(self, ABCSeries):
                 # GH#31182 Series._values returns EA, unpack for backward-compat
                 if getattr(self.dtype, "tz", None) is None:
+
+                    warnings.warn(
+                        f"Series.unique behavior with {self.dtype} dtype is "
+                        "deprecated. In a future version this will return a "
+                        f"{type(self._values)} instead of a np.ndarray",
+                        FutureWarning,
+                        stacklevel=find_stack_level(),
+                    )
                     result = np.asarray(result)
         else:
             result = unique1d(values)

pandas/tests/base/test_value_counts.py

@@ -193,6 +193,7 @@ def test_value_counts_bins(index_or_series):
 
 def test_value_counts_datetime64(index_or_series):
     klass = index_or_series
+    warn = None if klass is Index else FutureWarning
 
     # GH 3002, datetime64[ns]
     # don't test names though
@@ -223,12 +224,15 @@ def test_value_counts_datetime64(index_or_series):
         ["2010-01-01 00:00:00", "2009-01-01 00:00:00", "2008-09-09 00:00:00"],
         dtype="datetime64[ns]",
     )
+    with tm.assert_produces_warning(warn, match="DatetimeArray"):
+        unq = s.unique()
     if isinstance(s, Index):
-        tm.assert_index_equal(s.unique(), DatetimeIndex(expected))
+        tm.assert_index_equal(unq, DatetimeIndex(expected))
     else:
-        tm.assert_numpy_array_equal(s.unique(), expected)
+        tm.assert_numpy_array_equal(unq, expected)
 
-    assert s.nunique() == 3
+    with tm.assert_produces_warning(warn, match="DatetimeArray"):
+        assert s.nunique() == 3
 
     # with NaT
     s = df["dt"].copy()
@@ -243,7 +247,9 @@ def test_value_counts_datetime64(index_or_series):
     tm.assert_series_equal(result, expected_s)
 
     assert s.dtype == "datetime64[ns]"
-    unique = s.unique()
+    warn = None if isinstance(s, DatetimeIndex) else FutureWarning
+    with tm.assert_produces_warning(warn, match="DatetimeArray"):
+        unique = s.unique()
     assert unique.dtype == "datetime64[ns]"
 
     # numpy_array_equal cannot compare pd.NaT
@@ -254,8 +260,9 @@ def test_value_counts_datetime64(index_or_series):
         tm.assert_numpy_array_equal(unique[:3], expected)
         assert pd.isna(unique[3])
 
-    assert s.nunique() == 3
-    assert s.nunique(dropna=False) == 4
+    with tm.assert_produces_warning(warn, match="DatetimeArray"):
+        assert s.nunique() == 3
+        assert s.nunique(dropna=False) == 4
 
     # timedelta64[ns]
     td = df.dt - df.dt + timedelta(1)
@@ -269,7 +276,9 @@ def test_value_counts_datetime64(index_or_series):
     if isinstance(td, Index):
         tm.assert_index_equal(td.unique(), expected)
     else:
-        tm.assert_numpy_array_equal(td.unique(), expected.values)
+        with tm.assert_produces_warning(FutureWarning, match="TimedeltaArray"):
+            res = td.unique()
+        tm.assert_numpy_array_equal(res, expected.values)
 
     td2 = timedelta(1) + (df.dt - df.dt)
     td2 = klass(td2, name="dt")

pandas/tests/frame/test_reductions.py

@@ -70,7 +70,9 @@ def assert_stat_op_calc(
     f = getattr(frame, opname)
 
     if check_dates:
-        expected_warning = FutureWarning if opname in ["mean", "median"] else None
+        expected_warning = (
+            FutureWarning if opname in ["mean", "median", "nunique"] else None
+        )
         df = DataFrame({"b": date_range("1/1/2001", periods=2)})
         with tm.assert_produces_warning(expected_warning):
             result = getattr(df, opname)()

pandas/tests/reshape/merge/test_merge.py

@@ -625,15 +625,19 @@ def test_merge_nosort(self):
             ],
         }
         df = DataFrame.from_dict(d)
-        var3 = df.var3.unique()
+        with tm.assert_produces_warning(FutureWarning, match="DatetimeArray"):
+            var3 = df.var3.unique()
         var3.sort()
         new = DataFrame.from_dict({"var3": var3, "var8": np.random.random(7)})
 
         result = df.merge(new, on="var3", sort=False)
         exp = merge(df, new, on="var3", sort=False)
         tm.assert_frame_equal(result, exp)
 
-        assert (df.var3.unique() == result.var3.unique()).all()
+        with tm.assert_produces_warning(FutureWarning, match="DatetimeArray"):
+            res = df.var3.unique()
+            expected = result.var3.unique()
+        assert (res == expected).all()
 
     @pytest.mark.parametrize(
         ("sort", "values"), [(False, [1, 1, 0, 1, 1]), (True, [0, 1, 1, 1, 1])]

pandas/tests/test_algos.py

@@ -534,7 +534,13 @@ def test_dtype_preservation(self, any_numpy_dtype):
             data = [1, 2, 2]
             uniques = [1, 2]
 
-        result = Series(data, dtype=any_numpy_dtype).unique()
+        warn = None
+        if np.dtype(any_numpy_dtype).kind in ["m", "M"]:
+            warn = FutureWarning
+
+        ser = Series(data, dtype=any_numpy_dtype)
+        with tm.assert_produces_warning(warn, match="DatetimeArray|TimedeltaArray"):
+            result = ser.unique()
         expected = np.array(uniques, dtype=any_numpy_dtype)
 
         tm.assert_numpy_array_equal(result, expected)