Numpy 1.18 support

ci/azure/install.yml

@@ -16,9 +16,9 @@ steps:
         --pre \
         --upgrade \
         matplotlib \
+        numpy \
         pandas \
         scipy
-        # numpy \  # FIXME https://github.com/pydata/xarray/issues/3409
     pip install \
         --no-deps \
         --upgrade \

ci/requirements/py36.yml

@@ -25,7 +25,7 @@ dependencies:
   - nc-time-axis
   - netcdf4
   - numba
-  - numpy<1.18  # FIXME https://github.com/pydata/xarray/issues/3409
+  - numpy
   - pandas
   - pint
   - pip

ci/requirements/py37.yml

@@ -25,7 +25,7 @@ dependencies:
   - nc-time-axis
   - netcdf4
   - numba
-  - numpy<1.18  # FIXME https://github.com/pydata/xarray/issues/3409
+  - numpy
   - pandas
   - pint
   - pip

doc/whats-new.rst

@@ -115,6 +115,12 @@ Bug fixes
   (:issue:`3402`). By `Deepak Cherian <https://github.com/dcherian/>`_
 - Allow appending datetime and bool data variables to zarr stores.
   (:issue:`3480`). By `Akihiro Matsukawa <https://github.com/amatsukawa/>`_.
+- Add support for numpy >=1.18 (); bugfix mean() on datetime64 arrays on dask backend
+  (:issue:`3409`, :pull:`3537`). By `Guido Imperiale <https://github.com/crusaderky>`_.
+- Add support for pandas >=0.26 (:issue:`3440`).
+  By `Deepak Cherian <https://github.com/dcherian>`_.
+- Add support for pseudonetcdf >=3.1 (:pull:`3485`).
+  By `Barron Henderson <https://github.com/barronh>`_.
 
 Documentation
 ~~~~~~~~~~~~~
@@ -133,7 +139,6 @@ Documentation
 
 Internal Changes
 ~~~~~~~~~~~~~~~~
-
 - Added integration tests against `pint <https://pint.readthedocs.io/>`_.
   (:pull:`3238`, :pull:`3447`, :pull:`3493`, :pull:`3508`)
   by `Justus Magin <https://github.com/keewis>`_.

xarray/core/duck_array_ops.py

@@ -351,6 +351,32 @@ def f(values, axis=None, skipna=None, **kwargs):
 _mean = _create_nan_agg_method("mean")
 
 
+def _datetime_nanmin(array):
+    """nanmin() function for datetime64.
+
+    Caveats that this function deals with:
+
+    - In numpy < 1.18, min() on datetime64 incorrectly ignores NaT
+    - numpy nanmin() don't work on datetime64 (all versions at the moment of writing)
+    - dask min() does not work on datetime64 (all versions at the moment of writing)
+    """
+    from .dataarray import DataArray
+    from .variable import Variable
+
+    if isinstance(array, (DataArray, Variable)):
+        array = array.data
+
+    assert array.dtype.kind in "mM"
+    dtype = array.dtype
+    # (NaT).astype(float) does not produce NaN...
+    array = where(pandas_isnull(array), np.nan, array.astype(float))
+    array = min(array, skipna=True)
+    if isinstance(array, float):
+        array = np.array(array)
+    # ...but (NaN).astype("M8") does produce NaT
+    return array.astype(dtype)
+
+
 def datetime_to_numeric(array, offset=None, datetime_unit=None, dtype=float):
     """Convert an array containing datetime-like data to an array of floats.
 
@@ -370,7 +396,10 @@ def datetime_to_numeric(array, offset=None, datetime_unit=None, dtype=float):
     """
     # TODO: make this function dask-compatible?
     if offset is None:
-        offset = array.min()
+        if array.dtype.kind in "Mm":
+            offset = _datetime_nanmin(array)
+        else:
+            offset = min(array)
     array = array - offset
 
     if not hasattr(array, "dtype"):  # scalar is converted to 0d-array
@@ -401,7 +430,8 @@ def mean(array, axis=None, skipna=None, **kwargs):
 
     array = asarray(array)
     if array.dtype.kind in "Mm":
-        offset = min(array)
+        offset = _datetime_nanmin(array)
+
         # xarray always uses np.datetime64[ns] for np.datetime64 data
         dtype = "timedelta64[ns]"
         return (

xarray/tests/test_duck_array_ops.py

@@ -274,23 +274,39 @@ def assert_dask_array(da, dask):
 
 
 @arm_xfail
-@pytest.mark.parametrize("dask", [False, True])
-def test_datetime_reduce(dask):
-    time = np.array(pd.date_range("15/12/1999", periods=11))
-    time[8:11] = np.nan
-    da = DataArray(np.linspace(0, 365, num=11), dims="time", coords={"time": time})
-
-    if dask and has_dask:
-        chunks = {"time": 5}
-        da = da.chunk(chunks)
-
-    actual = da["time"].mean()
-    assert not pd.isnull(actual)
-    actual = da["time"].mean(skipna=False)
-    assert pd.isnull(actual)
-
-    # test for a 0d array
-    assert da["time"][0].mean() == da["time"][:1].mean()
+@pytest.mark.parametrize("dask", [False, True] if has_dask else [False])
+def test_datetime_mean(dask):
+    # Note: only testing numpy, as dask is broken upstream
+    da = DataArray(
+        np.array(["2010-01-01", "NaT", "2010-01-03", "NaT", "NaT"], dtype="M8"),
+        dims=["time"],
+    )
+    if dask:
+        # Trigger use case where a chunk is full of NaT
+        da = da.chunk({"time": 3})
+
+    expect = DataArray(np.array("2010-01-02", dtype="M8"))
+    expect_nat = DataArray(np.array("NaT", dtype="M8"))
+
+    actual = da.mean()
+    if dask:
+        assert actual.chunks is not None
+    assert_equal(actual, expect)
+
+    actual = da.mean(skipna=False)
+    if dask:
+        assert actual.chunks is not None
+    assert_equal(actual, expect_nat)
+
+    # tests for 1d array full of NaT
+    assert_equal(da[[1]].mean(), expect_nat)
+    assert_equal(da[[1]].mean(skipna=False), expect_nat)
+
+    # tests for a 0d array
+    assert_equal(da[0].mean(), da[0])
+    assert_equal(da[0].mean(skipna=False), da[0])
+    assert_equal(da[1].mean(), expect_nat)
+    assert_equal(da[1].mean(skipna=False), expect_nat)
 
 
 @requires_cftime