ENH: Add IntervalDtype support to IntervalIndex.astype (pandas-dev#19231)

Author: jschendel

doc/source/whatsnew/v0.23.0.txt

@@ -202,6 +202,7 @@ Other Enhancements
 - ``Resampler`` objects now have a functioning :attr:`~pandas.core.resample.Resampler.pipe` method.
   Previously, calls to ``pipe`` were diverted to  the ``mean`` method (:issue:`17905`).
 - :func:`~pandas.api.types.is_scalar` now returns ``True`` for ``DateOffset`` objects (:issue:`18943`).
+- ``IntervalIndex.astype`` now supports conversions between subtypes when passed an ``IntervalDtype`` (:issue:`19197`)
 
 .. _whatsnew_0230.api_breaking:
 

pandas/core/dtypes/dtypes.py

@@ -710,7 +710,8 @@ def __eq__(self, other):
             # None should match any subtype
             return True
         else:
-            return self.subtype == other.subtype
+            from pandas.core.dtypes.common import is_dtype_equal
+            return is_dtype_equal(self.subtype, other.subtype)
 
     @classmethod
     def is_dtype(cls, dtype):

pandas/core/indexes/interval.py

@@ -20,7 +20,8 @@
     is_scalar,
     is_float,
     is_number,
-    is_integer)
+    is_integer,
+    pandas_dtype)
 from pandas.core.indexes.base import (
     Index, _ensure_index,
     default_pprint, _index_shared_docs)
@@ -699,8 +700,16 @@ def copy(self, deep=False, name=None):
 
     @Appender(_index_shared_docs['astype'])
     def astype(self, dtype, copy=True):
-        if is_interval_dtype(dtype):
-            return self.copy() if copy else self
+        dtype = pandas_dtype(dtype)
+        if is_interval_dtype(dtype) and dtype != self.dtype:
+            try:
+                new_left = self.left.astype(dtype.subtype)
+                new_right = self.right.astype(dtype.subtype)
+            except TypeError:
+                msg = ('Cannot convert {dtype} to {new_dtype}; subtypes are '
+                       'incompatible')
+                raise TypeError(msg.format(dtype=self.dtype, new_dtype=dtype))
+            return self._shallow_copy(new_left, new_right)
         return super(IntervalIndex, self).astype(dtype, copy=copy)
 
     @cache_readonly

pandas/tests/dtypes/test_dtypes.py

@@ -534,6 +534,12 @@ def test_equality(self):
         assert not is_dtype_equal(IntervalDtype('int64'),
                                   IntervalDtype('float64'))
 
+        # invalid subtype comparisons do not raise when directly compared
+        dtype1 = IntervalDtype('float64')
+        dtype2 = IntervalDtype('datetime64[ns, US/Eastern]')
+        assert dtype1 != dtype2
+        assert dtype2 != dtype1
+
     @pytest.mark.parametrize('subtype', [
         None, 'interval', 'Interval', 'int64', 'uint64', 'float64',
         'complex128', 'datetime64', 'timedelta64', PeriodDtype('Q')])

pandas/tests/indexes/interval/test_astype.py

@@ -0,0 +1,209 @@
+from __future__ import division
+
+import pytest
+import numpy as np
+from pandas import (
+    Index,
+    IntervalIndex,
+    interval_range,
+    CategoricalIndex,
+    Timestamp,
+    Timedelta,
+    NaT)
+from pandas.core.dtypes.dtypes import CategoricalDtype, IntervalDtype
+import pandas.util.testing as tm
+
+
+class Base(object):
+    """Tests common to IntervalIndex with any subtype"""
+
+    def test_astype_idempotent(self, index):
+        result = index.astype('interval')
+        tm.assert_index_equal(result, index)
+
+        result = index.astype(index.dtype)
+        tm.assert_index_equal(result, index)
+
+    def test_astype_object(self, index):
+        result = index.astype(object)
+        expected = Index(index.values, dtype='object')
+        tm.assert_index_equal(result, expected)
+        assert not result.equals(index)
+
+    def test_astype_category(self, index):
+        result = index.astype('category')
+        expected = CategoricalIndex(index.values)
+        tm.assert_index_equal(result, expected)
+
+        result = index.astype(CategoricalDtype())
+        tm.assert_index_equal(result, expected)
+
+        # non-default params
+        categories = index.dropna().unique().values[:-1]
+        dtype = CategoricalDtype(categories=categories, ordered=True)
+        result = index.astype(dtype)
+        expected = CategoricalIndex(
+            index.values, categories=categories, ordered=True)
+        tm.assert_index_equal(result, expected)
+
+    @pytest.mark.parametrize('dtype', [
+        'int64', 'uint64', 'float64', 'complex128', 'period[M]',
+        'timedelta64', 'timedelta64[ns]', 'datetime64', 'datetime64[ns]',
+        'datetime64[ns, US/Eastern]'])
+    def test_astype_cannot_cast(self, index, dtype):
+        msg = 'Cannot cast IntervalIndex to dtype'
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype(dtype)
+
+    def test_astype_invalid_dtype(self, index):
+        msg = 'data type "fake_dtype" not understood'
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype('fake_dtype')
+
+
+class TestIntSubtype(Base):
+    """Tests specific to IntervalIndex with integer-like subtype"""
+
+    indexes = [
+        IntervalIndex.from_breaks(np.arange(-10, 11, dtype='int64')),
+        IntervalIndex.from_breaks(
+            np.arange(100, dtype='uint64'), closed='left'),
+    ]
+
+    @pytest.fixture(params=indexes)
+    def index(self, request):
+        return request.param
+
+    @pytest.mark.parametrize('subtype', [
+        'float64', 'datetime64[ns]', 'timedelta64[ns]'])
+    def test_subtype_conversion(self, index, subtype):
+        dtype = IntervalDtype(subtype)
+        result = index.astype(dtype)
+        expected = IntervalIndex.from_arrays(index.left.astype(subtype),
+                                             index.right.astype(subtype),
+                                             closed=index.closed)
+        tm.assert_index_equal(result, expected)
+
+    @pytest.mark.parametrize('subtype_start, subtype_end', [
+        ('int64', 'uint64'), ('uint64', 'int64')])
+    def test_subtype_integer(self, subtype_start, subtype_end):
+        index = IntervalIndex.from_breaks(np.arange(100, dtype=subtype_start))
+        dtype = IntervalDtype(subtype_end)
+        result = index.astype(dtype)
+        expected = IntervalIndex.from_arrays(index.left.astype(subtype_end),
+                                             index.right.astype(subtype_end),
+                                             closed=index.closed)
+        tm.assert_index_equal(result, expected)
+
+    @pytest.mark.xfail(reason='GH 15832')
+    def test_subtype_integer_errors(self):
+        # int64 -> uint64 fails with negative values
+        index = interval_range(-10, 10)
+        dtype = IntervalDtype('uint64')
+        with pytest.raises(ValueError):
+            index.astype(dtype)
+
+
+class TestFloatSubtype(Base):
+    """Tests specific to IntervalIndex with float subtype"""
+
+    indexes = [
+        interval_range(-10.0, 10.0, closed='neither'),
+        IntervalIndex.from_arrays([-1.5, np.nan, 0., 0., 1.5],
+                                  [-0.5, np.nan, 1., 1., 3.],
+                                  closed='both'),
+    ]
+
+    @pytest.fixture(params=indexes)
+    def index(self, request):
+        return request.param
+
+    @pytest.mark.parametrize('subtype', ['int64', 'uint64'])
+    def test_subtype_integer(self, subtype):
+        index = interval_range(0.0, 10.0)
+        dtype = IntervalDtype(subtype)
+        result = index.astype(dtype)
+        expected = IntervalIndex.from_arrays(index.left.astype(subtype),
+                                             index.right.astype(subtype),
+                                             closed=index.closed)
+        tm.assert_index_equal(result, expected)
+
+        # raises with NA
+        msg = 'Cannot convert NA to integer'
+        with tm.assert_raises_regex(ValueError, msg):
+            index.insert(0, np.nan).astype(dtype)
+
+    @pytest.mark.xfail(reason='GH 15832')
+    def test_subtype_integer_errors(self):
+        # float64 -> uint64 fails with negative values
+        index = interval_range(-10.0, 10.0)
+        dtype = IntervalDtype('uint64')
+        with pytest.raises(ValueError):
+            index.astype(dtype)
+
+        # float64 -> integer-like fails with non-integer valued floats
+        index = interval_range(0.0, 10.0, freq=0.25)
+        dtype = IntervalDtype('int64')
+        with pytest.raises(ValueError):
+            index.astype(dtype)
+
+        dtype = IntervalDtype('uint64')
+        with pytest.raises(ValueError):
+            index.astype(dtype)
+
+    @pytest.mark.parametrize('subtype', ['datetime64[ns]', 'timedelta64[ns]'])
+    def test_subtype_datetimelike(self, index, subtype):
+        dtype = IntervalDtype(subtype)
+        msg = 'Cannot convert .* to .*; subtypes are incompatible'
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype(dtype)
+
+
+class TestDatetimelikeSubtype(Base):
+    """Tests specific to IntervalIndex with datetime-like subtype"""
+
+    indexes = [
+        interval_range(Timestamp('2018-01-01'), periods=10, closed='neither'),
+        interval_range(Timestamp('2018-01-01'), periods=10).insert(2, NaT),
+        interval_range(Timestamp('2018-01-01', tz='US/Eastern'), periods=10),
+        interval_range(Timedelta('0 days'), periods=10, closed='both'),
+        interval_range(Timedelta('0 days'), periods=10).insert(2, NaT),
+    ]
+
+    @pytest.fixture(params=indexes)
+    def index(self, request):
+        return request.param
+
+    @pytest.mark.parametrize('subtype', ['int64', 'uint64'])
+    def test_subtype_integer(self, index, subtype):
+        dtype = IntervalDtype(subtype)
+        result = index.astype(dtype)
+        expected = IntervalIndex.from_arrays(index.left.astype(subtype),
+                                             index.right.astype(subtype),
+                                             closed=index.closed)
+        tm.assert_index_equal(result, expected)
+
+    def test_subtype_float(self, index):
+        dtype = IntervalDtype('float64')
+        msg = 'Cannot convert .* to .*; subtypes are incompatible'
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype(dtype)
+
+    def test_subtype_datetimelike(self):
+        # datetime -> timedelta raises
+        dtype = IntervalDtype('timedelta64[ns]')
+        msg = 'Cannot convert .* to .*; subtypes are incompatible'
+
+        index = interval_range(Timestamp('2018-01-01'), periods=10)
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype(dtype)
+
+        index = interval_range(Timestamp('2018-01-01', tz='CET'), periods=10)
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype(dtype)
+
+        # timedelta -> datetime raises
+        dtype = IntervalDtype('datetime64[ns]')
+        index = interval_range(Timedelta('0 days'), periods=10)
+        with tm.assert_raises_regex(TypeError, msg):
+            index.astype(dtype)

pandas/tests/indexes/interval/test_interval.py

@@ -415,26 +415,6 @@ def test_equals(self, closed):
                 np.arange(5), closed=other_closed)
             assert not expected.equals(expected_other_closed)
 
-    def test_astype(self, closed):
-        idx = self.create_index(closed=closed)
-        result = idx.astype(object)
-        tm.assert_index_equal(result, Index(idx.values, dtype='object'))
-        assert not idx.equals(result)
-        assert idx.equals(IntervalIndex.from_intervals(result))
-
-        result = idx.astype('interval')
-        tm.assert_index_equal(result, idx)
-        assert result.equals(idx)
-
-    @pytest.mark.parametrize('dtype', [
-        np.int64, np.float64, 'period[M]', 'timedelta64', 'datetime64[ns]',
-        'datetime64[ns, US/Eastern]'])
-    def test_astype_errors(self, closed, dtype):
-        idx = self.create_index(closed=closed)
-        msg = 'Cannot cast IntervalIndex to dtype'
-        with tm.assert_raises_regex(TypeError, msg):
-            idx.astype(dtype)
-
     @pytest.mark.parametrize('klass', [list, tuple, np.array, pd.Series])
     def test_where(self, closed, klass):
         idx = self.create_index(closed=closed)