sel with categorical index

doc/whats-new.rst

@@ -25,6 +25,8 @@ Breaking changes
 
 New Features
 ~~~~~~~~~~~~
+- Now :py:func:`sel` works dataarray / dataset with pd.CategoricalIndex. (:issue:`3669`)
+  By `Keisuke Fujii <https://github.com/fujiisoup>`_.
 - Implement :py:func:`median` and :py:func:`nanmedian` for dask arrays. This works by rechunking
   to a single chunk along all reduction axes. (:issue:`2999`).
   By `Deepak Cherian <https://github.com/dcherian>`_.

xarray/core/indexing.py

@@ -175,6 +175,10 @@ def convert_label_indexer(index, label, index_name="", method=None, tolerance=No
         if label.ndim == 0:
             if isinstance(index, pd.MultiIndex):
                 indexer, new_index = index.get_loc_level(label.item(), level=0)
+            elif isinstance(index, pd.CategoricalIndex):
+                if tolerance is not None:
+                    raise ValueError("tolerance is not valid with categorical index.")
+                indexer = index.get_loc(label.item())
             else:
                 indexer = index.get_loc(
                     label.item(), method=method, tolerance=tolerance

xarray/tests/test_dataset.py

@@ -1392,6 +1392,14 @@ def test_sel_dataarray_mindex(self):
                 )
             )
 
+    def test_sel_categorical(self):
+        ind = pd.Series(["foo", "bar"], dtype="category")
+        df = pd.DataFrame({"ind": ind, "values": [1, 2]})
+        ds = df.set_index("ind").to_xarray()
+        actual = ds.sel(ind="bar")
+        expected = ds.isel(ind=1)
+        assert_identical(expected, actual)
+
     def test_sel_drop(self):
         data = Dataset({"foo": ("x", [1, 2, 3])}, {"x": [0, 1, 2]})
         expected = Dataset({"foo": 1})