Use ddof in numbagg>=0.7.0 for aggregations (#8624)

* Use numbagg ddof for aggregations

(not yet for rolling)

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

Author: max-sixty

xarray/core/nputils.py

@@ -185,8 +185,12 @@ def f(values, axis=None, **kwargs):
             and pycompat.mod_version("numbagg") >= Version("0.5.0")
             and OPTIONS["use_numbagg"]
             and isinstance(values, np.ndarray)
-            # numbagg uses ddof=1 only, but numpy uses ddof=0 by default
-            and (("var" in name or "std" in name) and kwargs.get("ddof", 0) == 1)
+            # numbagg<0.7.0 uses ddof=1 only, but numpy uses ddof=0 by default
+            and (
+                pycompat.mod_version("numbagg") >= Version("0.7.0")
+                or ("var" not in name and "std" not in name)
+                or kwargs.get("ddof", 0) == 1
+            )
             # TODO: bool?
             and values.dtype.kind in "uifc"
             # and values.dtype.isnative
@@ -196,9 +200,12 @@ def f(values, axis=None, **kwargs):
 
             nba_func = getattr(numbagg, name, None)
             if nba_func is not None:
-                # numbagg does not take care dtype, ddof
+                # numbagg does not use dtype
                 kwargs.pop("dtype", None)
-                kwargs.pop("ddof", None)
+                # prior to 0.7.0, numbagg did not support ddof; we ensure it's limited
+                # to ddof=1 above.
+                if pycompat.mod_version("numbagg") < Version("0.7.0"):
+                    kwargs.pop("ddof", None)
                 return nba_func(values, axis=axis, **kwargs)
         if (
             _BOTTLENECK_AVAILABLE

xarray/tests/conftest.py

@@ -4,6 +4,7 @@
 import pandas as pd
 import pytest
 
+import xarray as xr
 from xarray import DataArray, Dataset
 from xarray.tests import create_test_data, requires_dask
 
@@ -13,6 +14,19 @@ def backend(request):
     return request.param
 
 
+@pytest.fixture(params=["numbagg", "bottleneck"])
+def compute_backend(request):
+    if request.param == "bottleneck":
+        options = dict(use_bottleneck=True, use_numbagg=False)
+    elif request.param == "numbagg":
+        options = dict(use_bottleneck=False, use_numbagg=True)
+    else:
+        raise ValueError
+
+    with xr.set_options(**options):
+        yield request.param
+
+
 @pytest.fixture(params=[1])
 def ds(request, backend):
     if request.param == 1:

xarray/tests/test_rolling.py

@@ -23,19 +23,6 @@
 ]
 
 
-@pytest.fixture(params=["numbagg", "bottleneck"])
-def compute_backend(request):
-    if request.param == "bottleneck":
-        options = dict(use_bottleneck=True, use_numbagg=False)
-    elif request.param == "numbagg":
-        options = dict(use_bottleneck=False, use_numbagg=True)
-    else:
-        raise ValueError
-
-    with xr.set_options(**options):
-        yield request.param
-
-
 @pytest.mark.parametrize("func", ["mean", "sum"])
 @pytest.mark.parametrize("min_periods", [1, 10])
 def test_cumulative(d, func, min_periods) -> None:

xarray/tests/test_variable.py

@@ -1754,7 +1754,8 @@ def test_reduce(self):
             v.mean(dim="x", axis=0)
 
     @requires_bottleneck
-    def test_reduce_use_bottleneck(self, monkeypatch):
+    @pytest.mark.parametrize("compute_backend", ["bottleneck"], indirect=True)
+    def test_reduce_use_bottleneck(self, monkeypatch, compute_backend):
         def raise_if_called(*args, **kwargs):
             raise RuntimeError("should not have been called")