Correctly accept a pandas.CategoricalDtype(pandas.IntervalDtype(...), ...) type (#17604)

From an offline discussion, a pandas object with an `category[interval[...]]` type would be incorrectly be interpreted as a `category[struct[...]]` type. This can cause further problems with `cudf.pandas` as a `category[struct[...]]` type cannot be properly interpreted by pandas.

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #17604

Author: mroeschke

python/cudf/cudf/core/column/categorical.py

@@ -1095,17 +1095,22 @@ def as_categorical_column(self, dtype: Dtype) -> Self:
             raise ValueError("dtype must be CategoricalDtype")
 
         if not isinstance(self.categories, type(dtype.categories._column)):
-            # If both categories are of different Column types,
-            # return a column full of Nulls.
-            codes = cast(
-                cudf.core.column.numerical.NumericalColumn,
-                column.as_column(
-                    _DEFAULT_CATEGORICAL_VALUE,
-                    length=self.size,
-                    dtype=self.codes.dtype,
-                ),
-            )
-            codes = as_unsigned_codes(len(dtype.categories), codes)
+            if isinstance(
+                self.categories.dtype, cudf.StructDtype
+            ) and isinstance(dtype.categories.dtype, cudf.IntervalDtype):
+                codes = self.codes
+            else:
+                # Otherwise if both categories are of different Column types,
+                # return a column full of nulls.
+                codes = cast(
+                    cudf.core.column.numerical.NumericalColumn,
+                    column.as_column(
+                        _DEFAULT_CATEGORICAL_VALUE,
+                        length=self.size,
+                        dtype=self.codes.dtype,
+                    ),
+                )
+                codes = as_unsigned_codes(len(dtype.categories), codes)
             return type(self)(
                 data=self.data,  # type: ignore[arg-type]
                 size=self.size,

python/cudf/cudf/core/column/column.py

@@ -2076,18 +2076,26 @@ def as_column(
             if isinstance(arbitrary.dtype, pd.DatetimeTZDtype):
                 new_tz = get_compatible_timezone(arbitrary.dtype)
                 arbitrary = arbitrary.astype(new_tz)
-            if isinstance(arbitrary.dtype, pd.CategoricalDtype) and isinstance(
-                arbitrary.dtype.categories.dtype, pd.DatetimeTZDtype
-            ):
-                new_tz = get_compatible_timezone(
-                    arbitrary.dtype.categories.dtype
-                )
-                new_cats = arbitrary.dtype.categories.astype(new_tz)
-                new_dtype = pd.CategoricalDtype(
-                    categories=new_cats, ordered=arbitrary.dtype.ordered
-                )
-                arbitrary = arbitrary.astype(new_dtype)
-
+            if isinstance(arbitrary.dtype, pd.CategoricalDtype):
+                if isinstance(
+                    arbitrary.dtype.categories.dtype, pd.DatetimeTZDtype
+                ):
+                    new_tz = get_compatible_timezone(
+                        arbitrary.dtype.categories.dtype
+                    )
+                    new_cats = arbitrary.dtype.categories.astype(new_tz)
+                    new_dtype = pd.CategoricalDtype(
+                        categories=new_cats, ordered=arbitrary.dtype.ordered
+                    )
+                    arbitrary = arbitrary.astype(new_dtype)
+                elif (
+                    isinstance(
+                        arbitrary.dtype.categories.dtype, pd.IntervalDtype
+                    )
+                    and dtype is None
+                ):
+                    # Conversion to arrow converts IntervalDtype to StructDtype
+                    dtype = cudf.CategoricalDtype.from_pandas(arbitrary.dtype)
             return as_column(
                 pa.array(arbitrary, from_pandas=True),
                 nan_as_null=nan_as_null,

python/cudf/cudf/tests/test_categorical.py

@@ -950,3 +950,13 @@ def test_index_set_categories(ordered):
     expected = pd_ci.set_categories([1, 2, 3, 4], ordered=ordered)
     result = cudf_ci.set_categories([1, 2, 3, 4], ordered=ordered)
     assert_eq(result, expected)
+
+
+def test_categorical_interval_pandas_roundtrip():
+    expected = cudf.Series(cudf.interval_range(0, 5)).astype("category")
+    result = cudf.Series.from_pandas(expected.to_pandas())
+    assert_eq(result, expected)
+
+    expected = pd.Series(pd.interval_range(0, 5)).astype("category")
+    result = cudf.Series.from_pandas(expected).to_pandas()
+    assert_eq(result, expected)