[FIX] Further fixes for CV with no df (#1548)

Author: marcopeix

neuralforecast/core.py

@@ -1863,6 +1863,11 @@ def _no_refit_cross_validation(
                     )
                 )
             else:
+                id_col, time_col, target_col = (
+                    self.id_col,
+                    self.time_col,
+                    self.target_col,
+                )
                 if verbose:
                     print("Using stored dataset.")
 
@@ -1953,13 +1958,14 @@ def _no_refit_cross_validation(
             if df is None:
                 # Reconstruct the target from the stored dataset. The dataset's
                 # temporal values are scaled, so undo any target scaling.
-                target_values = (
-                    self.dataset.temporal[:, self.dataset.y_idx].clone().numpy()
-                )
+                target_column = self.dataset.temporal[:, self.dataset.y_idx]
                 if self.scalers_:
                     target_values = self._scalers_target_inverse_transform(
-                        target_values.reshape(-1, 1), self.dataset.indptr
+                        target_column.clone().numpy().reshape(-1, 1),
+                        self.dataset.indptr,
                     ).reshape(-1)
+                else:
+                    target_values = target_column.numpy()
                 df = type(fcsts_df)(
                     {
                         id_col: ufp.repeat(self.uids, np.diff(self.dataset.indptr)),

tests/test_core.py

@@ -239,10 +239,11 @@ def test_neural_forecast_fit_cross_validation(setup_airplane_data):
     pd.testing.assert_frame_equal(after_fcst, init_fcst)
 
 
-# cross_validation() with no `df` should reuse the stored dataset
+# cross_validation() with no `df` should reuse the stored dataset.
+@pytest.mark.parametrize("local_scaler_type", [None, "standard"])
 @pytest.mark.parametrize("use_polars", [False, True])
 def test_cross_validation_without_df_uses_stored_dataset(
-    use_polars, setup_airplane_data, setup_airplane_data_polars
+    use_polars, local_scaler_type, setup_airplane_data, setup_airplane_data_polars
 ):
     if use_polars:
         df, _ = setup_airplane_data_polars
@@ -256,11 +257,12 @@ def test_cross_validation_without_df_uses_stored_dataset(
         assert_frame_equal = pd.testing.assert_frame_equal
 
     models = [NHITS(h=12, input_size=24, max_steps=2, random_seed=0)]
-    nf = NeuralForecast(models=models, freq=freq)
+    nf = NeuralForecast(models=models, freq=freq, local_scaler_type=local_scaler_type)
     nf.fit(df, **col_kwargs)
     # use_init_models resets to the same seeded weights before each run
     cv_with_df = nf.cross_validation(df, use_init_models=True, **col_kwargs)
-    cv_no_df = nf.cross_validation(use_init_models=True, **col_kwargs)
+    # No column kwargs here: the stored dataset's column names must be preserved.
+    cv_no_df = nf.cross_validation(use_init_models=True)
     assert_frame_equal(cv_no_df, cv_with_df)