[ENH] Partially replace pd.Int64Index with pd.Index

sktime/forecasting/base/_fh.py

@@ -22,8 +22,8 @@
 )
 from sktime.utils.validation.series import VALID_INDEX_TYPES
 
-RELATIVE_TYPES = (pd.Int64Index, pd.RangeIndex, pd.TimedeltaIndex)
-ABSOLUTE_TYPES = (pd.Int64Index, pd.RangeIndex, pd.DatetimeIndex, pd.PeriodIndex)
+RELATIVE_TYPES = (pd.RangeIndex, pd.TimedeltaIndex)
+ABSOLUTE_TYPES = (pd.RangeIndex, pd.DatetimeIndex, pd.PeriodIndex)
 assert set(RELATIVE_TYPES).issubset(VALID_INDEX_TYPES)
 assert set(ABSOLUTE_TYPES).issubset(VALID_INDEX_TYPES)
 VALID_FORECASTING_HORIZON_TYPES = (int, list, np.ndarray, pd.Index)
@@ -95,19 +95,22 @@ def _check_values(values: Union[VALID_FORECASTING_HORIZON_TYPES]) -> pd.Index:
     # isinstance() does not work here, because index types inherit from each
     # other,
     # hence we check for type equality here
-    if type(values) in VALID_INDEX_TYPES:
+    if (type(values) in VALID_INDEX_TYPES) or (
+        isinstance(values, pd.Index) and values.is_numeric()
+    ):
         pass
 
-    # convert single integer to pandas index, no further checks needed
+    # convert single integer or timedelta or dateoffset
+    # to pandas index, no further checks needed
     elif is_int(values):
-        return pd.Int64Index([values], dtype=int)
+        values = pd.Index([values], dtype=int)
 
     elif is_timedelta_or_date_offset(values):
-        return pd.Index([values])
+        values = pd.Index([values])
 
     # convert np.array or list to pandas index
     elif is_array(values) and array_is_int(values):
-        values = pd.Int64Index(values, dtype=int)
+        values = pd.Index(values, dtype=int)
 
     elif is_array(values) and array_is_timedelta_or_date_offset(values):
         values = pd.Index(values)
@@ -180,18 +183,21 @@ def __init__(
         # check types, note that isinstance() does not work here because index
         # types inherit from each other, hence we check for type equality
         error_msg = f"`values` type is not compatible with `is_relative={is_relative}`."
+        values_in_relative_types = type(values) in RELATIVE_TYPES
+        values_in_absolute_types = type(values) in ABSOLUTE_TYPES
+        values_is_numeric_index = isinstance(values, pd.Index) and values.is_numeric()
         if is_relative is None:
-            if type(values) in RELATIVE_TYPES:
+            if values_in_relative_types or values_is_numeric_index:
                 is_relative = True
-            elif type(values) in ABSOLUTE_TYPES:
+            elif values_in_absolute_types or values_is_numeric_index:
                 is_relative = False
             else:
                 raise TypeError(f"{type(values)} is not a supported fh index type")
         if is_relative:
-            if not type(values) in RELATIVE_TYPES:
+            if not (values_in_relative_types or values_is_numeric_index):
                 raise TypeError(error_msg)
         else:
-            if not type(values) in ABSOLUTE_TYPES:
+            if not (values_in_absolute_types or values_is_numeric_index):
                 raise TypeError(error_msg)
 
         self._values = values
@@ -262,7 +268,7 @@ def to_numpy(self, **kwargs) -> np.ndarray:
     # We cache the results from `to_relative()` and `to_absolute()` calls to speed up
     # computations, as these are the basic methods and often required internally when
     # calling different methods.
-    @lru_cache(typed=True)
+    @lru_cache(typed=True)  # noqa: B019
     def to_relative(self, cutoff=None):
         """Return forecasting horizon values relative to a cutoff.
 
@@ -319,7 +325,7 @@ def to_relative(self, cutoff=None):
 
             return self._new(relative, is_relative=True)
 
-    @lru_cache(typed=True)
+    @lru_cache(typed=True)  # noqa: B019
     def to_absolute(self, cutoff):
         """Return absolute version of forecasting horizon values.
 

sktime/forecasting/base/adapters/_statsmodels.py

@@ -47,7 +47,7 @@ def _fit(self, y, X=None, fh=None):
         """
         # statsmodels does not support the pd.Int64Index as required,
         # so we coerce them here to pd.RangeIndex
-        if isinstance(y, pd.Series) and type(y.index) == pd.Int64Index:
+        if isinstance(y, pd.Series) and y.index.is_numeric():
             y, X = _coerce_int_to_range_index(y, X)
         self._fit_forecaster(y, X)
         return self
@@ -113,7 +113,7 @@ def _coerce_int_to_range_index(y, X=None):
         np.testing.assert_array_equal(y.index, new_index)
     except AssertionError:
         raise ValueError(
-            "Coercion of pd.Int64Index to pd.RangeIndex "
+            "Coercion of integer pd.Index to pd.RangeIndex "
             "failed. Please provide `y_train` with a "
             "pd.RangeIndex."
         )

sktime/forecasting/base/tests/test_fh.py

@@ -64,7 +64,7 @@ def test_fh(index_type, fh_type, is_relative, steps):
     # get expected outputs
     if isinstance(steps, int):
         steps = np.array([steps])
-    fh_relative = pd.Int64Index(steps).sort_values()
+    fh_relative = pd.Index(steps).sort_values()
     fh_absolute = y.index[np.where(y.index == cutoff)[0] + steps].sort_values()
     fh_indexer = fh_relative - 1
     fh_oos = fh.to_pandas()[fh_relative > 0]
@@ -139,7 +139,7 @@ def test_check_fh_values_duplicate_input_values(arg):
 
 
 GOOD_ABSOLUTE_INPUT_ARGS = (
-    pd.Int64Index([1, 2, 3]),
+    pd.Index([1, 2, 3]),
     pd.period_range("2000-01-01", periods=3, freq="D"),
     pd.date_range("2000-01-01", periods=3, freq="M"),
     np.array([1, 2, 3]),
@@ -152,7 +152,9 @@ def test_check_fh_values_duplicate_input_values(arg):
 def test_check_fh_absolute_values_input_conversion_to_pandas_index(arg):
     """Test conversion of absolute horizons to pandas index."""
     output = ForecastingHorizon(arg, is_relative=False).to_pandas()
-    assert type(output) in VALID_INDEX_TYPES
+    assert (type(output) in VALID_INDEX_TYPES) or (
+        isinstance(output, pd.Index) and output.is_numeric()
+    )
 
 
 GOOD_RELATIVE_INPUT_ARGS = [
@@ -212,7 +214,7 @@ def test_coerce_duration_to_int(duration):
     ret = _coerce_duration_to_int(duration, freq=_get_freq(duration))
 
     # check output type is always integer
-    assert type(ret) in (pd.Int64Index, np.integer, int)
+    assert (type(ret) in (np.integer, int)) or ret.is_numeric()
 
     # check result
     if isinstance(duration, pd.Index):

sktime/forecasting/tests/_config.py

@@ -107,7 +107,7 @@
 ]
 
 INDEX_TYPE_LOOKUP = {
-    "int": pd.Int64Index,
+    "int": pd.Index,
     "range": pd.RangeIndex,
     "datetime": pd.DatetimeIndex,
     "period": pd.PeriodIndex,

sktime/utils/_testing/series.py

@@ -58,7 +58,7 @@ def _make_index(n_timepoints, index_type=None):
 
     elif index_type == "int":
         start = 3
-        return pd.Int64Index(np.arange(start, start + n_timepoints))
+        return pd.Index(np.arange(start, start + n_timepoints), dtype=int)
 
     else:
         raise ValueError(f"index_class: {index_type} is not supported")

sktime/utils/datetime.py

@@ -37,7 +37,7 @@ def _coerce_duration_to_int(duration, freq=None):
         duration[0], pd.tseries.offsets.BaseOffset
     ):
         count = _get_intervals_count_and_unit(freq)[0]
-        return pd.Int64Index([d.n / count for d in duration])
+        return pd.Index([d.n / count for d in duration], dtype=int)
     elif isinstance(duration, (pd.Timedelta, pd.TimedeltaIndex)):
         count, unit = _get_intervals_count_and_unit(freq)
         # integer conversion only works reliably with non-ambiguous units (
@@ -100,7 +100,7 @@ def _shift(x, by=1):
         Shifted time point
     """
     assert isinstance(x, (pd.Period, pd.Timestamp, int, np.integer)), type(x)
-    assert isinstance(by, (int, np.integer, pd.Int64Index)), type(by)
+    assert isinstance(by, (int, np.integer)) or by.is_numeric(), type(by)
     if isinstance(x, pd.Timestamp):
         if not hasattr(x, "freq") or x.freq is None:
             raise ValueError("No `freq` information available")

sktime/utils/validation/series.py

@@ -18,13 +18,7 @@
 
 # We currently support the following types for input data and time index types.
 VALID_DATA_TYPES = (pd.DataFrame, pd.Series, np.ndarray)
-VALID_INDEX_TYPES = (
-    pd.Int64Index,
-    pd.RangeIndex,
-    pd.PeriodIndex,
-    pd.DatetimeIndex,
-    pd.TimedeltaIndex,
-)
+VALID_INDEX_TYPES = (pd.RangeIndex, pd.PeriodIndex, pd.DatetimeIndex, pd.TimedeltaIndex)
 
 
 def _check_is_univariate(y, var_name="input"):
@@ -185,7 +179,7 @@ def check_time_index(
 
     # We here check for type equality because isinstance does not
     # work reliably because index types inherit from each other.
-    if not type(index) in VALID_INDEX_TYPES:
+    if not ((type(index) in VALID_INDEX_TYPES) or index.is_numeric()):
         raise NotImplementedError(
             f"{type(index)} is not supported for {var_name}, use "
             f"one of {VALID_INDEX_TYPES} instead."
@@ -194,7 +188,7 @@ def check_time_index(
     if enforce_index_type and type(index) is not enforce_index_type:
         raise NotImplementedError(
             f"{type(index)} is not supported for {var_name}, use "
-            f"type: {enforce_index_type} instead."
+            f"type: {enforce_index_type} or integer pd.Index instead."
         )
 
     # Check time index is ordered in time

sktime/utils/validation/tests/test_forecasting.py

@@ -12,7 +12,7 @@
 
 from sktime.utils.validation.forecasting import check_fh
 
-empty_input = (np.array([]), [], pd.Int64Index([]))
+empty_input = (np.array([], dtype=int), [], pd.Index([], dtype=int))
 
 
 @pytest.mark.parametrize("arg", empty_input)