Add dtype support for reduce methods.

doc/whats-new.rst

@@ -27,7 +27,9 @@ Documentation
 
 Enhancements
 ~~~~~~~~~~~~
-
+- reduce methods such as :py:func:`DataArray.sum()` now accepts ``dtype``
+  arguments. (:issue:`1838`)
+  By `Keisuke Fujii <https://github.com/fujiisoup>`_.
 - :py:func:`~plot.contourf()` learned to contour 2D variables that have both a
   1D co-ordinate (e.g. time) and a 2D co-ordinate (e.g. depth as a function of
   time) (:issue:`1737`).

xarray/core/duck_array_ops.py

@@ -175,11 +175,11 @@ def _create_nan_agg_method(name, numeric_only=False, np_compat=False,
                            no_bottleneck=False, coerce_strings=False,
                            keep_dims=False):
     def f(values, axis=None, skipna=None, **kwargs):
-        # ignore keyword args inserted by np.mean and other numpy aggregators
-        # automatically:
-        kwargs.pop('dtype', None)
-        kwargs.pop('out', None)
+        if kwargs.pop('out', None) is not None:
+            raise ValueError('`out` is not valid for {}'.format(name))
 
+        # If dtype is supplied, we use numpy's method.
+        dtype = kwargs.get('dtype', None)
         values = asarray(values)
 
         if coerce_strings and values.dtype.kind in 'SU':
@@ -192,14 +192,16 @@ def f(values, axis=None, skipna=None, **kwargs):
                     % (name, values.dtype))
             nanname = 'nan' + name
             if (isinstance(axis, tuple) or not values.dtype.isnative or
-                    no_bottleneck):
+                    no_bottleneck or
+                    (dtype is not None and np.dtype(dtype) != values.dtype)):
                 # bottleneck can't handle multiple axis arguments or non-native
                 # endianness
                 if np_compat:
                     eager_module = npcompat
                 else:
                     eager_module = np
             else:
+                kwargs.pop('dtype', None)
                 eager_module = bn
             func = _dask_or_eager_func(nanname, eager_module)
             using_numpy_nan_func = (eager_module is np or

xarray/tests/test_dataarray.py

@@ -1784,6 +1784,24 @@ def test_reduce(self):
                              dims=['x', 'y']).mean('x')
         assert_equal(actual, expected)
 
+    def test_reduce_dtype(self):
+        coords = {'x': [-1, -2], 'y': ['ab', 'cd', 'ef'],
+                  'lat': (['x', 'y'], [[1, 2, 3], [-1, -2, -3]]),
+                  'c': -999}
+        orig = DataArray([[-1, 0, 1], [-3, 0, 3]], coords, dims=['x', 'y'])
+
+        for dtype in [np.float32, np.float64, np.float128]:
+            assert orig.astype(float).mean(dtype=dtype).dtype == dtype
+
+    def test_reduce_out(self):
+        coords = {'x': [-1, -2], 'y': ['ab', 'cd', 'ef'],
+                  'lat': (['x', 'y'], [[1, 2, 3], [-1, -2, -3]]),
+                  'c': -999}
+        orig = DataArray([[-1, 0, 1], [-3, 0, 3]], coords, dims=['x', 'y'])
+
+        with pytest.raises(ValueError):
+            orig.mean(out=np.ones(orig.shape))
+
     # skip due to bug in older versions of numpy.nanpercentile
     def test_quantile(self):
         for q in [0.25, [0.50], [0.25, 0.75]]: