[MNT] 0.25.0 deprecations and change actions (#5613)

Carries out deprecations and change actions scheduled for 0.25.0:

* remove deprecated `forecasting.model_selection._split` module
* in forecasting `evaluate`, change order of return data frame
* in forecasting `evaluate`, remove `compute` arg. `kwargs` are still
accepted, with warning message. Bumped removal to 0.26.0 for smoother
transition.
* bump checks

README.md

@@ -169,8 +169,8 @@ flexible choice of soft dependencies is unavailable via `conda`)
 ``` python
 from sktime.datasets import load_airline
 from sktime.forecasting.base import ForecastingHorizon
-from sktime.forecasting.model_selection import temporal_train_test_split
 from sktime.forecasting.theta import ThetaForecaster
+from sktime.split import temporal_train_test_split
 from sktime.performance_metrics.forecasting import mean_absolute_percentage_error
 
 y = load_airline()

docs/source/get_started.rst

@@ -75,9 +75,9 @@ Forecasting
 
     >>> from sktime.datasets import load_airline
     >>> from sktime.forecasting.base import ForecastingHorizon
-    >>> from sktime.forecasting.model_selection import temporal_train_test_split
     >>> from sktime.forecasting.theta import ThetaForecaster
     >>> from sktime.performance_metrics.forecasting import mean_absolute_percentage_error
+    >>> from sktime.split import temporal_train_test_split
 
     >>> y = load_airline()
     >>> y_train, y_test = temporal_train_test_split(y)

docs/source/user_guide/introduction.rst

@@ -31,8 +31,8 @@ can simply write:
     from sktime.datasets import load_airline
     from sktime.forecasting.compose import make_reduction
     from sklearn.ensemble import RandomForestRegressor
-    from sktime.forecasting.model_selection import temporal_train_test_split
     from sktime.performance_metrics.forecasting import MeanAbsolutePercentageError
+    from sktime.split import temporal_train_test_split
 
     y = load_airline()
     y_train, y_test = temporal_train_test_split(y)

sktime/classification/early_classification/_probability_threshold.py

@@ -20,7 +20,7 @@
 from sktime.utils.validation.panel import check_X
 
 
-# TODO: fix this in 0.25.0
+# TODO: fix this in 0.26.0
 # base class should have been changed to BaseEarlyClassifier
 class ProbabilityThresholdEarlyClassifier(BaseClassifier):
     """Probability Threshold Early Classifier.

sktime/forecasting/all/__init__.py

@@ -11,13 +11,6 @@
 from sktime.datasets import load_airline, load_longley, load_lynx, load_shampoo_sales
 from sktime.forecasting.base import ForecastingHorizon
 from sktime.forecasting.model_evaluation import evaluate
-from sktime.forecasting.model_selection import (
-    CutoffSplitter,
-    ExpandingWindowSplitter,
-    SingleWindowSplitter,
-    SlidingWindowSplitter,
-    temporal_train_test_split,
-)
 from sktime.performance_metrics.forecasting import (
     GeometricMeanRelativeAbsoluteError,
     GeometricMeanRelativeSquaredError,
@@ -58,6 +51,13 @@
     relative_loss,
 )
 from sktime.registry import all_estimators
+from sktime.split import (
+    CutoffSplitter,
+    ExpandingWindowSplitter,
+    SingleWindowSplitter,
+    SlidingWindowSplitter,
+    temporal_train_test_split,
+)
 from sktime.transformations.series.detrend import Deseasonalizer, Detrender
 from sktime.utils.plotting import plot_series
 

sktime/forecasting/base/_fh.py

@@ -806,7 +806,7 @@ def _to_relative(fh: ForecastingHorizon, cutoff=None) -> ForecastingHorizon:
             absolute = _coerce_to_period(absolute, freq=fh.freq)
             cutoff = _coerce_to_period(cutoff, freq=fh.freq)
 
-        # TODO: 0.25.0:
+        # TODO: 0.26.0:
         # Check at every minor release whether lower pandas bound >=0.15.0
         # if yes, can remove the workaround in the "else" condition and the check
         #

sktime/forecasting/compose/_pipeline.py

@@ -170,7 +170,7 @@ def _get_inverse_transform(self, transformers, y, X=None, mode=None):
                         if len(levels) == 1:
                             levels = levels[0]
                         yt[ix] = y.xs(ix, level=levels, axis=1)
-                        # todo 0.25.0 - check why this cannot be easily removed
+                        # todo 0.26.0 - check why this cannot be easily removed
                         # in theory, we should get rid of the "Coverage" case treatment
                         # (the legacy naming convention was removed in 0.23.0)
                         # deal with the "Coverage" case, we need to get rid of this

sktime/forecasting/compose/tests/test_pipeline.py

@@ -485,7 +485,7 @@ def test_forecastx_logic():
 )
 def test_forecastx_fit_behavior():
     from sktime.forecasting.compose import ForecastX
-    from sktime.forecasting.model_selection import temporal_train_test_split
+    from sktime.split import temporal_train_test_split
 
     y, X = load_longley()
     y_train, y_test, X_train, X_test = temporal_train_test_split(y, X)

sktime/forecasting/model_evaluation/_functions.py

@@ -318,8 +318,6 @@ def _evaluate_window(x, meta):
         return result
 
 
-# todo 0.25.0: remove compute argument and docstring
-# todo 0.25.0: remove kwargs and docstring
 def evaluate(
     forecaster,
     cv,
@@ -330,7 +328,6 @@ def evaluate(
     return_data: bool = False,
     error_score: Union[str, int, float] = np.nan,
     backend: Optional[str] = None,
-    compute: bool = None,
     cv_X=None,
     backend_params: Optional[dict] = None,
     **kwargs,
@@ -422,9 +419,6 @@ def evaluate(
         "threading" is unlikely to see speed ups due to the GIL and the serialization
         backend (``cloudpickle``) for "dask" and "loky" is generally more robust
         than the standard ``pickle`` library used in "multiprocessing".
-    compute : bool, default=True, deprecated and will be removed in 0.25.0.
-        If backend="dask", whether returned DataFrame is computed.
-        If set to True, returns `pd.DataFrame`, otherwise `dask.dataframe.DataFrame`.
     cv_X : sktime BaseSplitter descendant, optional
         determines split of ``X`` into test and train folds
         default is ``X`` being split to identical ``loc`` indices as ``y``
@@ -525,10 +519,10 @@ def evaluate(
                 "installed, but dask is not present in the python environment"
             )
 
-    # todo 0.25.0: remove kwargs and this warning
+    # todo 0.26.0: remove kwargs and this warning
     if kwargs != {}:
         warnings.warn(
-            "in evaluate, kwargs will no longer be supported from sktime 0.25.0. "
+            "in evaluate, kwargs are no longer supported. "
             "to pass configuration arguments to the parallelization backend, "
             "use backend_params instead. "
             f"The following kwargs were found: {kwargs.keys()}, pass these as "
@@ -537,28 +531,8 @@ def evaluate(
             stacklevel=2,
         )
 
-    # todo 0.25.0: remove compute argument and logic, and remove this warning
-    if compute is not None:
-        warnings.warn(
-            "the compute argument of evaluate is deprecated and will be removed "
-            "in sktime 0.25.0. For the same behaviour in the future, "
-            'use backend="dask_lazy"',
-            DeprecationWarning,
-            stacklevel=2,
-        )
-    if compute is None:
-        compute = True
-    if backend == "dask" and not compute:
-        backend = "dask_lazy"
-
     _check_strategy(strategy)
     cv = check_cv(cv, enforce_start_with_window=True)
-    # TODO: remove lines(four lines below) and 599-612 in v0.25.0
-    if isinstance(scoring, list):
-        raise_warn, num = True, len(scoring)
-    else:
-        raise_warn, num = False, 1
-    # removal until here
     scoring = _check_scores(scoring)
 
     ALLOWED_SCITYPES = ["Series", "Panel", "Hierarchical"]
@@ -669,21 +643,4 @@ def gen_y_X_train_test(y, X, cv, cv_X):
     # final formatting of results DataFrame
     results = results.reset_index(drop=True)
 
-    # TODO: remove 16 lines below and 451-455 in v0.25.0
-    if raise_warn:
-        warnings.warn(
-            "Starting v0.25.0 model_evaluation.evaluate module will rearrange "
-            "all metric columns to the left of its output result DataFrame. "
-            "Please use loc references when addressing the columns. You can "
-            "safely ignore this warning if you don't use evaluate function directly.",
-            DeprecationWarning,
-            stacklevel=2,
-        )
-        columns = results.columns.to_list()
-        non_first_metrics = []
-        for _ in range(1, num):
-            metric = columns.pop(1)
-            non_first_metrics.append(metric)
-        results = results.reindex(columns=columns + non_first_metrics)
-    #  removal until here
     return results

sktime/forecasting/model_selection/__init__.py

@@ -4,29 +4,11 @@
 
 __author__ = ["mloning", "kkoralturk"]
 __all__ = [
-    "CutoffSplitter",
-    "SameLocSplitter",
-    "SingleWindowSplitter",
-    "SlidingWindowSplitter",
-    "temporal_train_test_split",
-    "ExpandingGreedySplitter",
-    "ExpandingWindowSplitter",
-    "TestPlusTrainSplitter",
     "ForecastingGridSearchCV",
     "ForecastingRandomizedSearchCV",
     "ForecastingSkoptSearchCV",
 ]
 
-from sktime.forecasting.model_selection._split import (
-    CutoffSplitter,
-    ExpandingGreedySplitter,
-    ExpandingWindowSplitter,
-    SameLocSplitter,
-    SingleWindowSplitter,
-    SlidingWindowSplitter,
-    TestPlusTrainSplitter,
-    temporal_train_test_split,
-)
 from sktime.forecasting.model_selection._tune import (
     ForecastingGridSearchCV,
     ForecastingRandomizedSearchCV,

sktime/forecasting/model_selection/_tune.py

@@ -544,9 +544,7 @@ class ForecastingGridSearchCV(BaseGridSearch):
     --------
     >>> from sktime.datasets import load_shampoo_sales
     >>> from sktime.forecasting.model_selection import ForecastingGridSearchCV
-    >>> from sktime.split import (
-    ...     ExpandingWindowSplitter,
-    ...     ExpandingWindowSplitter)
+    >>> from sktime.split import ExpandingWindowSplitter
     >>> from sktime.forecasting.naive import NaiveForecaster
     >>> y = load_shampoo_sales()
     >>> fh = [1,2,3]

sktime/forecasting/trend/_pwl_trend_forecaster.py

@@ -60,8 +60,8 @@ class ProphetPiecewiseLinearTrendForecaster(_ProphetAdapter):
     --------
     >>> from sktime.datasets import load_airline
     >>> from sktime.forecasting.trend import ProphetPiecewiseLinearTrendForecaster
-    >>> from sktime.forecasting.model_selection import temporal_train_test_split
     >>> from sktime.forecasting.base import ForecastingHorizon
+    >>> from sktime.split import temporal_train_test_split
     >>> y =load_airline().to_timestamp(freq='M')
     >>> y_train, y_test = temporal_train_test_split(y)
     >>> fh = ForecastingHorizon(y.index, is_relative=False)

sktime/forecasting/trend/tests/test_pwl_trend.py

@@ -11,11 +11,11 @@
 
 from sktime.datasets import load_airline
 from sktime.forecasting.base import ForecastingHorizon
-from sktime.forecasting.model_selection import temporal_train_test_split
 from sktime.forecasting.trend import (
     PolynomialTrendForecaster,
     ProphetPiecewiseLinearTrendForecaster,
 )
+from sktime.split import temporal_train_test_split
 from sktime.tests.test_switch import run_test_for_class
 
 

sktime/split/sameloc.py

@@ -141,10 +141,7 @@ def get_test_params(cls, parameter_set="default"):
             `create_test_instance` uses the first (or only) dictionary in `params`
         """
         from sktime.datasets import load_airline
-        from sktime.forecasting.model_selection import (
-            ExpandingWindowSplitter,
-            SingleWindowSplitter,
-        )
+        from sktime.split import ExpandingWindowSplitter, SingleWindowSplitter
 
         y = load_airline()
         y_temp = y[:60]

sktime/split/temporal_train_test_split.py

@@ -91,8 +91,8 @@ def temporal_train_test_split(
 
     Examples
     --------
-    >>> from sktime.forecasting.model_selection import temporal_train_test_split
     >>> from sktime.datasets import load_airline, load_osuleaf
+    >>> from sktime.split import temporal_train_test_split
     >>> from sktime.utils._testing.panel import _make_panel
     >>> # univariate time series
     >>> y = load_airline()

sktime/split/testplustrain.py

@@ -131,10 +131,7 @@ def get_test_params(cls, parameter_set="default"):
             `MyClass(**params)` or `MyClass(**params[i])` creates a valid test instance.
             `create_test_instance` uses the first (or only) dictionary in `params`
         """
-        from sktime.forecasting.model_selection import (
-            ExpandingWindowSplitter,
-            SingleWindowSplitter,
-        )
+        from sktime.split import ExpandingWindowSplitter, SingleWindowSplitter
 
         cv_1 = ExpandingWindowSplitter(fh=[2, 4], initial_window=24, step_length=12)
         cv_2 = SingleWindowSplitter(fh=[2, 4], window_length=24)