Return Interval object in pandas compat mode for IntervalIndex reduct…

…ions (#16523)

xref #16507

In non pandas compat mode, I think this still makes sense to return a `dict` since that's the "scalar" type of a cudf struct/interval type, but in pandas compat mode we should match pandas and return an Interval.

Authors:
  - Matthew Roeschke (https://github.com/mroeschke)
  - GALI PREM SAGAR (https://github.com/galipremsagar)

Approvers:
  - GALI PREM SAGAR (https://github.com/galipremsagar)

URL: #16523

python/cudf/cudf/_lib/reduce.pyx

@@ -61,7 +61,11 @@ def reduce(reduction_op, Column incol, dtype=None, **kwargs):
             result,
             dtype=col_dtype.__class__(precision, scale),
         ).value
-    return DeviceScalar.from_pylibcudf(result).value
+    scalar = DeviceScalar.from_pylibcudf(result).value
+    if isinstance(col_dtype, cudf.StructDtype):
+        # TODO: Utilize column_metadata in libcudf to maintain field labels
+        return dict(zip(col_dtype.fields.keys(), scalar.values()))
+    return scalar
 
 
 @acquire_spill_lock()

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

@@ -11,6 +11,7 @@
 from cudf.core.dtypes import IntervalDtype
 
 if TYPE_CHECKING:
+    from cudf._typing import ScalarLike
     from cudf.core.column import ColumnBase
 
 
@@ -186,3 +187,16 @@ def element_indexing(self, index: int):
         if cudf.get_option("mode.pandas_compatible"):
             return pd.Interval(**result, closed=self.dtype.closed)
         return result
+
+    def _reduce(
+        self,
+        op: str,
+        skipna: bool | None = None,
+        min_count: int = 0,
+        *args,
+        **kwargs,
+    ) -> ScalarLike:
+        result = super()._reduce(op, skipna, min_count, *args, **kwargs)
+        if cudf.get_option("mode.pandas_compatible"):
+            return pd.Interval(**result, closed=self.dtype.closed)
+        return result

python/cudf/cudf/tests/test_interval.py

@@ -194,3 +194,14 @@ def test_intervaldtype_eq_string_with_attributes():
     dtype = cudf.IntervalDtype("int64", closed="left")
     assert dtype == "interval"
     assert dtype == "interval[int64, left]"
+
+
+def test_reduction_return_interval_pandas_compatible():
+    ii = pd.IntervalIndex.from_tuples(
+        [("2017-01-03", "2017-01-04")], dtype="interval[datetime64[ns], right]"
+    )
+    cudf_ii = cudf.IntervalIndex.from_pandas(ii)
+    with cudf.option_context("mode.pandas_compatible", True):
+        result = cudf_ii.min()
+    expected = ii.min()
+    assert result == expected