DEPR: Deprecate generic timestamp dtypes (#15987)

* DEPR: Deprecate generic timestamp dtypes

We only use the nanosecond frequency, and numpy
doesn't even handle generic timestamp dtypes well.

xref gh-15524 (comment).

* TST: Use pytest idioms in series/test_dtypes.py

Author: gfyoung

doc/source/whatsnew/v0.20.0.txt

@@ -1204,6 +1204,7 @@ Deprecations
 - ``SparseArray.to_dense()`` has deprecated the ``fill`` parameter, as that parameter was not being respected (:issue:`14647`)
 - ``SparseSeries.to_dense()`` has deprecated the ``sparse_only`` parameter (:issue:`14647`)
 - ``Series.repeat()`` has deprecated the ``reps`` parameter in favor of ``repeats`` (:issue:`12662`)
+- The ``Series`` constructor and ``.astype`` method have deprecated accepting timestamp dtypes without a frequency (e.g. ``np.datetime64``) for the ``dtype`` parameter (:issue:`15524`)
 - ``Index.repeat()`` and ``MultiIndex.repeat()`` have deprecated the ``n`` parameter in favor of ``repeats`` (:issue:`12662`)
 - ``Categorical.searchsorted()`` and ``Series.searchsorted()`` have deprecated the ``v`` parameter in favor of ``value`` (:issue:`12662`)
 - ``TimedeltaIndex.searchsorted()``, ``DatetimeIndex.searchsorted()``, and ``PeriodIndex.searchsorted()`` have deprecated the ``key`` parameter in favor of ``value`` (:issue:`12662`)

pandas/tests/series/test_constructors.py

@@ -839,3 +839,30 @@ def test_constructor_cast_object(self):
         s = Series(date_range('1/1/2000', periods=10), dtype=object)
         exp = Series(date_range('1/1/2000', periods=10))
         tm.assert_series_equal(s, exp)
+
+    def test_constructor_generic_timestamp_deprecated(self):
+        # see gh-15524
+
+        with tm.assert_produces_warning(FutureWarning):
+            dtype = np.timedelta64
+            s = Series([], dtype=dtype)
+
+            assert s.empty
+            assert s.dtype == 'm8[ns]'
+
+        with tm.assert_produces_warning(FutureWarning):
+            dtype = np.datetime64
+            s = Series([], dtype=dtype)
+
+            assert s.empty
+            assert s.dtype == 'M8[ns]'
+
+        # These timestamps have the wrong frequencies,
+        # so an Exception should be raised now.
+        msg = "cannot convert timedeltalike"
+        with tm.assertRaisesRegexp(TypeError, msg):
+            Series([], dtype='m8[ps]')
+
+        msg = "cannot convert datetimelike"
+        with tm.assertRaisesRegexp(TypeError, msg):
+            Series([], dtype='M8[ps]')

pandas/tests/series/test_dtypes.py

@@ -1,9 +1,13 @@
 # coding=utf-8
 # pylint: disable-msg=E1101,W0612
 
-import sys
+import pytest
+
 from datetime import datetime
+
+import sys
 import string
+import warnings
 
 from numpy import nan
 import numpy as np
@@ -12,152 +16,199 @@
 
 from pandas.compat import lrange, range, u
 from pandas import compat
-from pandas.util.testing import assert_series_equal
 import pandas.util.testing as tm
 
 from .common import TestData
 
 
-class TestSeriesDtypes(TestData, tm.TestCase):
+class TestSeriesDtypes(TestData):
 
-    def test_astype(self):
+    @pytest.mark.parametrize("dtype", ["float32", "float64",
+                                       "int64", "int32"])
+    def test_astype(self, dtype):
         s = Series(np.random.randn(5), name='foo')
+        as_typed = s.astype(dtype)
 
-        for dtype in ['float32', 'float64', 'int64', 'int32']:
-            astyped = s.astype(dtype)
-            self.assertEqual(astyped.dtype, dtype)
-            self.assertEqual(astyped.name, s.name)
+        assert as_typed.dtype == dtype
+        assert as_typed.name == s.name
 
     def test_dtype(self):
 
-        self.assertEqual(self.ts.dtype, np.dtype('float64'))
-        self.assertEqual(self.ts.dtypes, np.dtype('float64'))
-        self.assertEqual(self.ts.ftype, 'float64:dense')
-        self.assertEqual(self.ts.ftypes, 'float64:dense')
-        assert_series_equal(self.ts.get_dtype_counts(), Series(1, ['float64']))
-        assert_series_equal(self.ts.get_ftype_counts(), Series(
-            1, ['float64:dense']))
-
-    def test_astype_cast_nan_inf_int(self):
-        # GH14265, check nan and inf raise error when converting to int
-        types = [np.int32, np.int64]
-        values = [np.nan, np.inf]
+        assert self.ts.dtype == np.dtype('float64')
+        assert self.ts.dtypes == np.dtype('float64')
+        assert self.ts.ftype == 'float64:dense'
+        assert self.ts.ftypes == 'float64:dense'
+        tm.assert_series_equal(self.ts.get_dtype_counts(),
+                               Series(1, ['float64']))
+        tm.assert_series_equal(self.ts.get_ftype_counts(),
+                               Series(1, ['float64:dense']))
+
+    @pytest.mark.parametrize("value", [np.nan, np.inf])
+    @pytest.mark.parametrize("dtype", [np.int32, np.int64])
+    def test_astype_cast_nan_inf_int(self, dtype, value):
+        # gh-14265: check NaN and inf raise error when converting to int
         msg = 'Cannot convert non-finite values \\(NA or inf\\) to integer'
+        s = Series([value])
 
-        for this_type in types:
-            for this_val in values:
-                s = Series([this_val])
-                with self.assertRaisesRegexp(ValueError, msg):
-                    s.astype(this_type)
+        with tm.assertRaisesRegexp(ValueError, msg):
+            s.astype(dtype)
 
-    def test_astype_cast_object_int(self):
+    @pytest.mark.parametrize("dtype", [int, np.int8, np.int64])
+    def test_astype_cast_object_int_fail(self, dtype):
         arr = Series(["car", "house", "tree", "1"])
+        with pytest.raises(ValueError):
+            arr.astype(dtype)
 
-        self.assertRaises(ValueError, arr.astype, int)
-        self.assertRaises(ValueError, arr.astype, np.int64)
-        self.assertRaises(ValueError, arr.astype, np.int8)
-
+    def test_astype_cast_object_int(self):
         arr = Series(['1', '2', '3', '4'], dtype=object)
         result = arr.astype(int)
-        self.assert_series_equal(result, Series(np.arange(1, 5)))
+
+        tm.assert_series_equal(result, Series(np.arange(1, 5)))
 
     def test_astype_datetimes(self):
         import pandas._libs.tslib as tslib
-
         s = Series(tslib.iNaT, dtype='M8[ns]', index=lrange(5))
+
         s = s.astype('O')
-        self.assertEqual(s.dtype, np.object_)
+        assert s.dtype == np.object_
 
         s = Series([datetime(2001, 1, 2, 0, 0)])
+
         s = s.astype('O')
-        self.assertEqual(s.dtype, np.object_)
+        assert s.dtype == np.object_
 
         s = Series([datetime(2001, 1, 2, 0, 0) for i in range(3)])
+
         s[1] = np.nan
-        self.assertEqual(s.dtype, 'M8[ns]')
-        s = s.astype('O')
-        self.assertEqual(s.dtype, np.object_)
+        assert s.dtype == 'M8[ns]'
 
-    def test_astype_str(self):
-        # GH4405
-        digits = string.digits
-        s1 = Series([digits * 10, tm.rands(63), tm.rands(64), tm.rands(1000)])
-        s2 = Series([digits * 10, tm.rands(63), tm.rands(64), nan, 1.0])
-        types = (compat.text_type, np.str_)
-        for typ in types:
-            for s in (s1, s2):
-                res = s.astype(typ)
-                expec = s.map(compat.text_type)
-                assert_series_equal(res, expec)
-
-        # GH9757
-        # Test str and unicode on python 2.x and just str on python 3.x
-        for tt in set([str, compat.text_type]):
-            ts = Series([Timestamp('2010-01-04 00:00:00')])
-            s = ts.astype(tt)
-            expected = Series([tt('2010-01-04')])
-            assert_series_equal(s, expected)
-
-            ts = Series([Timestamp('2010-01-04 00:00:00', tz='US/Eastern')])
-            s = ts.astype(tt)
-            expected = Series([tt('2010-01-04 00:00:00-05:00')])
-            assert_series_equal(s, expected)
-
-            td = Series([Timedelta(1, unit='d')])
-            s = td.astype(tt)
-            expected = Series([tt('1 days 00:00:00.000000000')])
-            assert_series_equal(s, expected)
+        s = s.astype('O')
+        assert s.dtype == np.object_
+
+    @pytest.mark.parametrize("dtype", [compat.text_type, np.str_])
+    @pytest.mark.parametrize("series", [Series([string.digits * 10,
+                                                tm.rands(63),
+                                                tm.rands(64),
+                                                tm.rands(1000)]),
+                                        Series([string.digits * 10,
+                                                tm.rands(63),
+                                                tm.rands(64), nan, 1.0])])
+    def test_astype_str_map(self, dtype, series):
+        # see gh-4405
+        result = series.astype(dtype)
+        expected = series.map(compat.text_type)
+        tm.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("dtype", [str, compat.text_type])
+    def test_astype_str_cast(self, dtype):
+        # see gh-9757: test str and unicode on python 2.x
+        # and just str on python 3.x
+        ts = Series([Timestamp('2010-01-04 00:00:00')])
+        s = ts.astype(dtype)
+
+        expected = Series([dtype('2010-01-04')])
+        tm.assert_series_equal(s, expected)
+
+        ts = Series([Timestamp('2010-01-04 00:00:00', tz='US/Eastern')])
+        s = ts.astype(dtype)
+
+        expected = Series([dtype('2010-01-04 00:00:00-05:00')])
+        tm.assert_series_equal(s, expected)
+
+        td = Series([Timedelta(1, unit='d')])
+        s = td.astype(dtype)
+
+        expected = Series([dtype('1 days 00:00:00.000000000')])
+        tm.assert_series_equal(s, expected)
 
     def test_astype_unicode(self):
-
-        # GH7758
-        # a bit of magic is required to set default encoding encoding to utf-8
+        # see gh-7758: A bit of magic is required to set
+        # default encoding to utf-8
         digits = string.digits
         test_series = [
             Series([digits * 10, tm.rands(63), tm.rands(64), tm.rands(1000)]),
             Series([u('データーサイエンス、お前はもう死んでいる')]),
-
         ]
 
         former_encoding = None
+
         if not compat.PY3:
-            # in python we can force the default encoding for this test
+            # In Python, we can force the default encoding for this test
             former_encoding = sys.getdefaultencoding()
             reload(sys)  # noqa
+
             sys.setdefaultencoding("utf-8")
         if sys.getdefaultencoding() == "utf-8":
             test_series.append(Series([u('野菜食べないとやばい')
                                        .encode("utf-8")]))
+
         for s in test_series:
             res = s.astype("unicode")
             expec = s.map(compat.text_type)
-            assert_series_equal(res, expec)
-        # restore the former encoding
+            tm.assert_series_equal(res, expec)
+
+        # Restore the former encoding
         if former_encoding is not None and former_encoding != "utf-8":
             reload(sys)  # noqa
             sys.setdefaultencoding(former_encoding)
 
     def test_astype_dict(self):
-        # GH7271
+        # see gh-7271
         s = Series(range(0, 10, 2), name='abc')
 
         result = s.astype({'abc': str})
         expected = Series(['0', '2', '4', '6', '8'], name='abc')
-        assert_series_equal(result, expected)
+        tm.assert_series_equal(result, expected)
 
         result = s.astype({'abc': 'float64'})
         expected = Series([0.0, 2.0, 4.0, 6.0, 8.0], dtype='float64',
                           name='abc')
-        assert_series_equal(result, expected)
-
-        self.assertRaises(KeyError, s.astype, {'abc': str, 'def': str})
-        self.assertRaises(KeyError, s.astype, {0: str})
-
-    def test_complexx(self):
-        # GH4819
-        # complex access for ndarray compat
+        tm.assert_series_equal(result, expected)
+
+        with pytest.raises(KeyError):
+            s.astype({'abc': str, 'def': str})
+
+        with pytest.raises(KeyError):
+            s.astype({0: str})
+
+    def test_astype_generic_timestamp_deprecated(self):
+        # see gh-15524
+        data = [1]
+
+        with tm.assert_produces_warning(FutureWarning,
+                                        check_stacklevel=False):
+            s = Series(data)
+            dtype = np.datetime64
+            result = s.astype(dtype)
+            expected = Series(data, dtype=dtype)
+            tm.assert_series_equal(result, expected)
+
+        with tm.assert_produces_warning(FutureWarning,
+                                        check_stacklevel=False):
+            s = Series(data)
+            dtype = np.timedelta64
+            result = s.astype(dtype)
+            expected = Series(data, dtype=dtype)
+            tm.assert_series_equal(result, expected)
+
+    @pytest.mark.parametrize("dtype", np.typecodes['All'])
+    def test_astype_empty_constructor_equality(self, dtype):
+        # see gh-15524
+
+        if dtype not in ('S', 'V'):  # poor support (if any) currently
+            with warnings.catch_warnings(record=True):
+                # Generic timestamp dtypes ('M' and 'm') are deprecated,
+                # but we test that already in series/test_constructors.py
+
+                init_empty = Series([], dtype=dtype)
+                as_type_empty = Series([]).astype(dtype)
+                tm.assert_series_equal(init_empty, as_type_empty)
+
+    def test_complex(self):
+        # see gh-4819: complex access for ndarray compat
         a = np.arange(5, dtype=np.float64)
         b = Series(a + 4j * a)
+
         tm.assert_numpy_array_equal(a, b.real)
         tm.assert_numpy_array_equal(4 * a, b.imag)
 
@@ -166,23 +217,22 @@ def test_complexx(self):
         tm.assert_numpy_array_equal(4 * a, b.imag)
 
     def test_arg_for_errors_in_astype(self):
-        # issue #14878
-
-        sr = Series([1, 2, 3])
+        # see gh-14878
+        s = Series([1, 2, 3])
 
-        with self.assertRaises(ValueError):
-            sr.astype(np.float64, errors=False)
+        with pytest.raises(ValueError):
+            s.astype(np.float64, errors=False)
 
         with tm.assert_produces_warning(FutureWarning):
-            sr.astype(np.int8, raise_on_error=True)
+            s.astype(np.int8, raise_on_error=True)
 
-        sr.astype(np.int8, errors='raise')
+        s.astype(np.int8, errors='raise')
 
     def test_intercept_astype_object(self):
         series = Series(date_range('1/1/2000', periods=10))
 
-        # this test no longer makes sense as series is by default already
-        # M8[ns]
+        # This test no longer makes sense, as
+        # Series is by default already M8[ns].
         expected = series.astype('object')
 
         df = DataFrame({'a': series,
@@ -192,9 +242,9 @@ def test_intercept_astype_object(self):
         tm.assert_series_equal(df.dtypes, exp_dtypes)
 
         result = df.values.squeeze()
-        self.assertTrue((result[:, 0] == expected.values).all())
+        assert (result[:, 0] == expected.values).all()
 
         df = DataFrame({'a': series, 'b': ['foo'] * len(series)})
 
         result = df.values.squeeze()
-        self.assertTrue((result[:, 0] == expected.values).all())
+        assert (result[:, 0] == expected.values).all()