Add FlagGrouper

Closes #472

cf_xarray/__init__.py

@@ -9,6 +9,6 @@
 from .options import set_options  # noqa
 from .utils import _get_version
 
-from . import geometry  # noqa
+from . import geometry, groupers  # noqa
 
 __version__ = _get_version()

cf_xarray/groupers.py

@@ -0,0 +1,27 @@
+import numpy as np
+import pandas as pd
+from xarray.groupers import EncodedGroups, Grouper
+
+
+class FlagGrouper(Grouper):
+    def factorize(self, group) -> EncodedGroups:
+        assert "flag_values" in group.attrs
+        assert "flag_meanings" in group.attrs
+
+        values = np.array(group.attrs["flag_values"])
+        full_index = pd.Index(group.attrs["flag_meanings"].split(" "))
+
+        if group.dtype.kind in "iu" and (np.diff(values) == 1).all():
+            # optimize
+            codes = group.data - group.data[0].astype(int)
+        else:
+            codes, _ = pd.factorize(group.data.ravel())
+
+        codes_da = group.copy(data=codes.reshape(group.shape))
+        codes_da.attrs.pop("flag_values")
+        codes_da.attrs.pop("flag_meanings")
+
+        return EncodedGroups(codes=codes_da, full_index=full_index)
+
+    def reset(self):
+        pass

cf_xarray/tests/test_groupers.py

@@ -0,0 +1,15 @@
+import numpy as np
+from xarray.testing import assert_identical
+
+from cf_xarray.datasets import flag_excl
+from cf_xarray.groupers import FlagGrouper
+
+
+def test_flag_grouper():
+    ds = flag_excl.to_dataset().set_coords("flag_var")
+    ds["foo"] = ("time", np.arange(8))
+    actual = ds.groupby(flag_var=FlagGrouper()).mean()
+    expected = ds.groupby("flag_var").mean()
+    expected["flag_var"] = ["flag_1", "flag_2", "flag_3"]
+    expected["flag_var"].attrs["standard_name"] = "flag_mutual_exclusive"
+    assert_identical(actual, expected)

doc/api.rst

@@ -21,13 +21,20 @@ Geometries
 ----------
 .. autosummary::
     :toctree: generated/
-
     geometry.decode_geometries
+
     geometry.encode_geometries
     geometry.shapely_to_cf
     geometry.cf_to_shapely
     geometry.GeometryNames
 
+
+Groupers
+--------
+.. autosummary::
+    :toctree: generated/
+    groupers.FlagGrouper
+
 .. currentmodule:: xarray
 
 DataArray

doc/flags.md

@@ -60,6 +60,31 @@ You can also check whether a DataArray has the appropriate attributes to be reco
 da.cf.is_flag_variable
 ```
 
+## GroupBy
+
+Flag variables, such as that above, are naturally used for GroupBy operations.
+cf-xarray provides a `FlagGrouper` that understands the `flag_meanings` and `flag_values` attributes.
+
+```{code-cell}
+import cf_xarray as cfxr
+import numpy as np
+
+from cf_xarray.datasets import flag_excl
+from cf_xarray.groupers import FlagGrouper
+
+ds = flag_excl.to_dataset().set_coords('flag_var')
+ds["foo"] = ("time", np.arange(8))
+ds.flag_var
+```
+
+The `flag_var` array has the need attributes.
+
+```{code-cell}
+ds.groupby(flag_var=FlagGrouper()).mean()
+```
+
+Note how the output coordinate has the values from `flag_meanings`
+
 ## Flag Masks
 
 ```{warning}