Handle non-numpy dtypes without erroring

Fixes GH716

Author: shoyer

doc/whats-new.rst

@@ -69,6 +69,9 @@ Bug fixes
 
 - ``DataArray.to_masked_array`` always returns masked array with mask being an array
 (not a scalar value) (:issue:`684`)
+- You can now pass pandas objects with non-numpy dtypes (e.g., ``categorical``
+  or ``datetime64`` with a timezone) into xray without an error
+  (:issue:`716`).
 
 v0.6.2 (unreleased)
 -------------------

xray/core/indexing.py

@@ -371,6 +371,10 @@ def __init__(self, array, dtype=None):
             # if a PeriodIndex, force an object dtype
             if isinstance(array, pd.PeriodIndex):
                 dtype = np.dtype('O')
+            elif hasattr(array, 'categories'):
+                dtype = array.categories.dtype
+            elif not utils.is_valid_numpy_dtype(array.dtype):
+                dtype = np.dtype('O')
             else:
                 dtype = array.dtype
         self._dtype = dtype
@@ -387,7 +391,7 @@ def __array__(self, dtype=None):
             with suppress(AttributeError):
                 # this might not be public API
                 array = array.asobject
-        return array.values.astype(dtype)
+        return np.asarray(array, dtype)
 
     def __getitem__(self, key):
         if isinstance(key, tuple) and len(key) == 1:

xray/core/utils.py

@@ -169,6 +169,15 @@ def is_scalar(value):
             value is None)
 
 
+def is_valid_numpy_dtype(dtype):
+    try:
+        np.dtype(dtype)
+    except (TypeError, ValueError):
+        return False
+    else:
+        return True
+
+
 def dict_equiv(first, second, compat=equivalent):
     """Test equivalence of two dict-like objects. If any of the values are
     numpy arrays, compare them correctly.

xray/test/test_variable.py

@@ -5,6 +5,7 @@
 
 from distutils.version import LooseVersion
 import numpy as np
+import pytz
 import pandas as pd
 
 from xray import Variable, Dataset, DataArray
@@ -153,7 +154,7 @@ def test_0d_time_data(self):
     def test_datetime64_conversion(self):
         times = pd.date_range('2000-01-01', periods=3)
         for values, preserve_source in [
-                (times, False),
+                (times, True),
                 (times.values, True),
                 (times.values.astype('datetime64[s]'), False),
                 (times.to_pydatetime(), False),
@@ -163,15 +164,12 @@ def test_datetime64_conversion(self):
             self.assertArrayEqual(v.values, times.values)
             self.assertEqual(v.values.dtype, np.dtype('datetime64[ns]'))
             same_source = source_ndarray(v.values) is source_ndarray(values)
-            if preserve_source and self.cls is Variable:
-                self.assertTrue(same_source)
-            else:
-                self.assertFalse(same_source)
+            assert preserve_source == same_source
 
     def test_timedelta64_conversion(self):
         times = pd.timedelta_range(start=0, periods=3)
         for values, preserve_source in [
-                (times, False),
+                (times, True),
                 (times.values, True),
                 (times.values.astype('timedelta64[s]'), False),
                 (times.to_pytimedelta(), False),
@@ -181,10 +179,7 @@ def test_timedelta64_conversion(self):
             self.assertArrayEqual(v.values, times.values)
             self.assertEqual(v.values.dtype, np.dtype('timedelta64[ns]'))
             same_source = source_ndarray(v.values) is source_ndarray(values)
-            if preserve_source and self.cls is Variable:
-                self.assertTrue(same_source)
-            else:
-                self.assertFalse(same_source)
+            assert preserve_source == same_source
 
     def test_object_conversion(self):
         data = np.arange(5).astype(str).astype(object)
@@ -405,6 +400,22 @@ def test_aggregate_complex(self):
         expected = Variable((), 0.5 + 1j)
         self.assertVariableAllClose(v.mean(), expected)
 
+    def test_pandas_cateogrical_dtype(self):
+        data = pd.Categorical(np.arange(10, dtype='int64'))
+        v = self.cls('x', data)
+        print(v)  # should not error
+        assert v.dtype == 'int64'
+
+    def test_pandas_datetime64_with_tz(self):
+        data = pd.date_range(start='2000-01-01',
+                             tz=pytz.timezone('America/New_York'),
+                             periods=10, freq='1h')
+        v = self.cls('x', data)
+        print(v)  # should not error
+        if 'America/New_York' in str(data.dtype):
+            # pandas is new enough that it has datetime64 with timezone dtype
+            assert v.dtype == 'object'
+
 
 class TestVariable(TestCase, VariableSubclassTestCases):
     cls = staticmethod(Variable)