BUG: IntervalIndex.factorize with non-nano (pandas-dev#56099)

* BUG: IntervalIndex.factorize with non-nano * GH ref

Author: jbrockmendel

doc/source/whatsnew/v2.2.0.rst

@@ -404,6 +404,7 @@ Strings
 Interval
 ^^^^^^^^
 - Bug in :class:`Interval` ``__repr__`` not displaying UTC offsets for :class:`Timestamp` bounds. Additionally the hour, minute and second components will now be shown. (:issue:`55015`)
+- Bug in :meth:`IntervalIndex.factorize` and :meth:`Series.factorize` with :class:`IntervalDtype` with datetime64 or timedelta64 intervals not preserving non-nanosecond units (:issue:`56099`)
 - Bug in :meth:`IntervalIndex.from_arrays` when passed ``datetime64`` or ``timedelta64`` arrays with mismatched resolutions constructing an invalid ``IntervalArray`` object (:issue:`55714`)
 - Bug in :meth:`IntervalIndex.get_indexer` with datetime or timedelta intervals incorrectly matching on integer targets (:issue:`47772`)
 - Bug in :meth:`IntervalIndex.get_indexer` with timezone-aware datetime intervals incorrectly matching on a sequence of timezone-naive targets (:issue:`47772`)

pandas/core/arrays/interval.py

@@ -391,12 +391,7 @@ def _from_sequence(
 
  @classmethod
  def _from_factorized(cls, values: np.ndarray, original: IntervalArray) -> Self:
- if len(values) == 0:
- # An empty array returns object-dtype here. We can't create
- # a new IA from an (empty) object-dtype array, so turn it into the
- # correct dtype.
- values = values.astype(original.dtype.subtype)
- return cls(values, closed=original.closed)
+ return cls._from_sequence(values, dtype=original.dtype)
 
  _interval_shared_docs["from_breaks"] = textwrap.dedent(
  """

pandas/tests/test_algos.py

@@ -536,6 +536,25 @@ def test_factorize_mixed_values(self, data, expected_codes, expected_uniques):
  tm.assert_numpy_array_equal(codes, expected_codes)
  tm.assert_index_equal(uniques, expected_uniques)
 
+ def test_factorize_interval_non_nano(self, unit):
+ # GH#56099
+ left = DatetimeIndex(["2016-01-01", np.nan, "2015-10-11"]).as_unit(unit)
+ right = DatetimeIndex(["2016-01-02", np.nan, "2015-10-15"]).as_unit(unit)
+ idx = IntervalIndex.from_arrays(left, right)
+ codes, cats = idx.factorize()
+ assert cats.dtype == f"interval[datetime64[{unit}], right]"
+
+ ts = Timestamp(0).as_unit(unit)
+ idx2 = IntervalIndex.from_arrays(left - ts, right - ts)
+ codes2, cats2 = idx2.factorize()
+ assert cats2.dtype == f"interval[timedelta64[{unit}], right]"
+
+ idx3 = IntervalIndex.from_arrays(
+ left.tz_localize("US/Pacific"), right.tz_localize("US/Pacific")
+ )
+ codes3, cats3 = idx3.factorize()
+ assert cats3.dtype == f"interval[datetime64[{unit}, US/Pacific], right]"
+
 
 class TestUnique:
  def test_ints(self):