[ENH] in BaseForecaster, move check for capability:insample to _check_fh boilerplate

sktime/forecasting/base/_base.py

@@ -616,7 +616,7 @@ def predict_quantiles(self, fh=None, X=None, alpha=None):
         # input checks and conversions
 
         # check fh and coerce to ForecastingHorizon, if not already passed in fit
-        fh = self._check_fh(fh)
+        fh = self._check_fh(fh, pred_int=True)
 
         # default alpha
         if alpha is None:
@@ -702,7 +702,7 @@ def predict_interval(self, fh=None, X=None, coverage=0.90):
         # input checks and conversions
 
         # check fh and coerce to ForecastingHorizon, if not already passed in fit
-        fh = self._check_fh(fh)
+        fh = self._check_fh(fh, pred_int=True)
 
         # check alpha and coerce to list
         coverage = check_alpha(coverage, name="coverage")
@@ -789,7 +789,7 @@ def predict_var(self, fh=None, X=None, cov=False):
         # input checks and conversions
 
         # check fh and coerce to ForecastingHorizon, if not already passed in fit
-        fh = self._check_fh(fh)
+        fh = self._check_fh(fh, pred_int=True)
 
         # check and convert X
         X_inner = self._check_X(X=X)
@@ -861,7 +861,7 @@ def predict_proba(self, fh=None, X=None, marginal=True):
         # input checks and conversions
 
         # check fh and coerce to ForecastingHorizon, if not already passed in fit
-        fh = self._check_fh(fh)
+        fh = self._check_fh(fh, pred_int=True)
 
         # check and convert X
         X_inner = self._check_X(X=X)
@@ -1172,16 +1172,16 @@ def update_predict_single(
 
         self.check_is_fitted()
 
-        # check fh and coerce to ForecastingHorizon, if not already passed in fit
-        fh = self._check_fh(fh)
-
         # input checks and minor coercions on X, y
         X_inner, y_inner = self._check_X_y(X=X, y=y)
 
         # update internal _X/_y with the new X/y
         # this also updates cutoff from y
         self._update_y_X(y_inner, X_inner)
 
+        # check fh and coerce to ForecastingHorizon, if not already passed in fit
+        fh = self._check_fh(fh)
+
         # checks and conversions complete, pass to inner update_predict_single
         if not self._is_vectorized:
             y_pred = self._update_predict_single(
@@ -1771,7 +1771,7 @@ def fh(self):
 
         return self._fh
 
-    def _check_fh(self, fh):
+    def _check_fh(self, fh, pred_int=False):
         """Check, set and update the forecasting horizon.
 
         Called from all methods where fh can be passed:
@@ -1791,6 +1791,7 @@ def _check_fh(self, fh):
         Parameters
         ----------
         fh : None, int, list, np.ndarray or ForecastingHorizon
+        pred_int: Check pred_int:insample tag instead of insample tag.
 
         Returns
         -------
@@ -1878,6 +1879,23 @@ def _check_fh(self, fh):
                     "horizon, please re-fit the forecaster. " + msg
                 )
             # if existing one and new match, ignore new one
+        in_sample_pred = (
+            self.get_tag("capability:insample")
+            if not pred_int
+            else self.get_tag("capability:pred_int:insample")
+        )
+        if (
+            not in_sample_pred
+            and self._fh is not None
+            and not self._fh.is_all_out_of_sample(self._cutoff)
+        ):
+            msg = (
+                f"{self.__class__.__name__} "
+                f"can not perform in-sample prediction. "
+                f"Found fh with in sample index: "
+                f"{fh}"
+            )
+            raise NotImplementedError(msg)
 
         return self._fh
 

sktime/forecasting/base/adapters/_neuralforecast.py

@@ -266,9 +266,6 @@ def _fit(
         ValueError
             When ``freq="auto"`` and cannot be interpreted from ``ForecastingHorizon``
         """
-        if not fh.is_all_out_of_sample(cutoff=self.cutoff):
-            raise NotImplementedError("in-sample prediction is currently not supported")
-
         # A. freq is given {use this}
         # B. freq is auto
         #     B1. freq is infered from fh {use this}

sktime/forecasting/base/adapters/_pytorch.py

@@ -126,9 +126,6 @@ def _predict(self, X=None, fh=None):
             fh = self.fh
         fh = fh.to_relative(self.cutoff)
 
-        if min(fh._values) < 0:
-            raise NotImplementedError("LTSF is not supporting insample predictions.")
-
         if max(fh._values) > self.network.pred_len or min(fh._values) < 0:
             raise ValueError(
                 f"fh of {fh} passed to {self.__class__.__name__} is not "

sktime/forecasting/compose/_reduce.py

@@ -283,10 +283,7 @@ def _predict_in_sample(self, fh, X=None, **kwargs):
         # Note that we currently only support out-of-sample predictions. For the
         # direct and multioutput strategy, we need to check this already during fit,
         # as the fh is required for fitting.
-        raise NotImplementedError(
-            f"Generating in-sample predictions is not yet "
-            f"implemented for {self.__class__.__name__}."
-        )
+        pass
 
     @classmethod
     def get_test_params(cls, parameter_set="default"):
@@ -578,9 +575,6 @@ def _fit(self, y, X, fh):
                     + "observation size"
                 )
 
-        if not self.fh.is_all_out_of_sample(self.cutoff):
-            raise NotImplementedError("In-sample predictions are not implemented.")
-
         yt, Xt = self._transform(y, X)
         if hasattr(Xt, "columns"):
             Xt.columns = Xt.columns.astype(str)
@@ -783,9 +777,6 @@ def _fit(self, y, X, fh):
         # We currently only support out-of-sample predictions. For the direct
         # strategy, we need to check this at the beginning of fit, as the fh is
         # required for fitting.
-        if not self.fh.is_all_out_of_sample(self.cutoff):
-            raise NotImplementedError("In-sample predictions are not implemented.")
-
         self.window_length_ = check_window_length(
             self.window_length, n_timepoints=len(y)
         )
@@ -1125,9 +1116,6 @@ def _fit(self, y, X, fh):
         if X is not None:
             X = None
 
-        if len(self.fh.to_in_sample(self.cutoff)) > 0:
-            raise NotImplementedError("In-sample predictions are not implemented")
-
         self.window_length_ = check_window_length(
             self.window_length, n_timepoints=len(y)
         )

sktime/forecasting/hf_transformers_forecaster.py

@@ -248,11 +248,6 @@ def _predict(self, fh, X=None):
             fh = self.fh
         fh = fh.to_relative(self.cutoff)
 
-        if min(fh._values) < 0:
-            raise NotImplementedError(
-                "The huggingface adapter is not supporting insample predictions."
-            )
-
         self.model.eval()
         from torch import from_numpy
 

sktime/forecasting/tests/test_all_forecasters.py

@@ -227,8 +227,13 @@ def test_predict_time_index(
 
         # if estimator cannot forecast in-sample and fh is in-sample, terminate
         # if the tag correctly states this, we consider this fine as per contract
-        # todo: check that estimator raises error message when fitting instead
+        # check that estimator raises error message when fitting
         if not fh_is_oos and not estimator_instance.get_tag("capability:insample"):
+            with pytest.raises(
+                NotImplementedError,
+                match="can not perform in-sample prediction",
+            ):
+                estimator_instance.fit(y_train, fh=fh)
             return None
 
         estimator_instance.fit(y_train, fh=fh)
@@ -245,10 +250,16 @@ def test_predict_time_index(
 
         # if cannot forecast in-sample probabilistically, and fh is in-sample, terminate
         # if the tag correctly states this, we consider this fine as per contract
-        # todo: check that estimator raises error message when fitting instead
+        # check that estimator raises error message when fitting
         if not fh_is_oos:
             if not estimator_instance.get_tag("capability:pred_int:insample"):
-                return None
+                if estimator_instance.get_tag("capability:pred_int"):
+                    with pytest.raises(
+                        NotImplementedError,
+                        match="can not perform in-sample prediction",
+                    ):
+                        estimator_instance.predict_interval()
+                    return None
 
         if estimator_instance.get_tag("capability:pred_int"):
             y_pred_int = estimator_instance.predict_interval()

sktime/forecasting/tests/test_neuralforecast.py

@@ -25,9 +25,7 @@ def test_neural_forecast_univariate_y_without_X(model_class) -> None:
     model = model_class(freq="A-DEC", max_steps=5, trainer_kwargs={"logger": False})
 
     # attempt fit with negative fh
-    with pytest.raises(
-        NotImplementedError, match="in-sample prediction is currently not supported"
-    ):
+    with pytest.raises(NotImplementedError):
         model.fit(y_train, fh=[-2, -1, 0, 1, 2])
 
     # train model