BUG: groupby.groups with NA categories fails

Author: rhshadrach

pandas/core/groupby/grouper.py

@@ -12,11 +12,16 @@
 
 import numpy as np
 
+from pandas._libs import (
+    algos as libalgos,
+)
 from pandas._libs.tslibs import OutOfBoundsDatetime
 from pandas.errors import InvalidIndexError
 from pandas.util._decorators import cache_readonly
 
 from pandas.core.dtypes.common import (
+    ensure_int64,
+    ensure_platform_int,
     is_list_like,
     is_scalar,
 )
@@ -38,7 +43,10 @@
 )
 from pandas.core.series import Series
 
-from pandas.io.formats.printing import pprint_thing
+from pandas.io.formats.printing import (
+    PrettyDict,
+    pprint_thing,
+)
 
 if TYPE_CHECKING:
     from collections.abc import (
@@ -668,8 +676,16 @@ def _codes_and_uniques(self) -> tuple[npt.NDArray[np.signedinteger], ArrayLike]:
     def groups(self) -> dict[Hashable, Index]:
         codes, uniques = self._codes_and_uniques
         uniques = Index._with_infer(uniques, name=self.name)
-        cats = Categorical.from_codes(codes, uniques, validate=False)
-        return self._index.groupby(cats)
+
+        r, counts = libalgos.groupsort_indexer(ensure_platform_int(codes), len(uniques))
+        counts = ensure_int64(counts).cumsum()
+        _result = (r[start:end] for start, end in zip(counts, counts[1:]))
+        result = dict(zip(uniques, _result))
+
+        # map to the label
+        result = {k: self._index.take(v) for k, v in result.items()}
+
+        return PrettyDict(result)
 
     @property
     def observed_grouping(self) -> Grouping:

pandas/tests/groupby/test_categorical.py

@@ -506,6 +506,23 @@ def test_observed_groups(observed):
     tm.assert_dict_equal(result, expected)
 
 
+def test_groups_na_category(dropna, observed):
+    # https://github.com/pandas-dev/pandas/issues/61356
+    df = DataFrame(
+        {"cat": Categorical(["a", np.nan, "a"], categories=list("adb"))},
+        index=list("xyz"),
+    )
+    g = df.groupby("cat", observed=observed, dropna=dropna)
+
+    result = g.groups
+    expected = {"a": Index(["x", "z"])}
+    if not dropna:
+        expected |= {np.nan: Index(["y"])}
+    if not observed:
+        expected |= {"b": Index([]), "d": Index([])}
+    tm.assert_dict_equal(result, expected)
+
+
 @pytest.mark.parametrize(
     "keys, expected_values, expected_index_levels",
     [