Fix/categorical validation features (#2776)

* Format validation set

* Add cat_feature to validation set

* Test categorical validation set

* Update darts/tests/models/forecasting/test_regression_models.py

Co-authored-by: madtoinou <[email protected]>

---------

Co-authored-by: madtoinou <[email protected]>

Author: jonasblanc

darts/models/forecasting/catboost_model.py

@@ -374,6 +374,7 @@ def _add_val_set_to_kwargs(
                     data=val_set[0],
                     label=val_set[1],
                     weight=val_weights[i] if val_weights is not None else None,
+                    cat_features=self._categorical_indices,
                 )
             )
         kwargs[val_set_name] = val_pools

darts/models/forecasting/regression_model.py

@@ -677,6 +677,8 @@ def _create_lagged_data(
         ):
             sample_weights = sample_weights.ravel()
 
+        features, labels = self._format_samples(features, labels)
+
         return features, labels, sample_weights
 
     def _format_samples(
@@ -733,9 +735,7 @@ def _fit_model(
                     "`sample_weight` was ignored since underlying regression model's "
                     "`fit()` method does not support it."
                 )
-        training_samples, training_labels = self._format_samples(
-            training_samples, training_labels
-        )
+
         self.model.fit(
             training_samples, training_labels, **sample_weight_kwargs, **kwargs
         )

darts/tests/models/forecasting/test_regression_models.py

@@ -3596,6 +3596,10 @@ def test_categorical_features_passed_to_fit_correctly(self, model_cls_and_module
             future_covariates,
         ) = self.inputs_for_tests_categorical_covariates()
 
+        series_train, series_val = series.split_after(0.6)
+        past_cov_train, past_cov_val = past_covariates.split_after(0.6)
+        future_cov_train, future_cov_val = future_covariates.split_after(0.6)
+
         original_fit = model.model.fit
         intercepted_args = {}
 
@@ -3610,24 +3614,39 @@ def intercept_fit_args(*args, **kwargs):
             side_effect=intercept_fit_args,
         ):
             model.fit(
-                series=series,
-                past_covariates=past_covariates,
-                future_covariates=future_covariates,
+                series=series_train,
+                past_covariates=past_cov_train,
+                future_covariates=future_cov_train,
+                val_series=series_val,
+                val_past_covariates=past_cov_val,
+                val_future_covariates=future_cov_val,
             )
 
             expected_cat_indices = [2, 3, 5]
             cat_param_name = model._categorical_fit_param
+            eval_set_param_name, _ = model.val_set_params
             if model_cls == CatBoostModel:
                 model_cat_indices = model.model.get_cat_feature_indices()
                 kwargs_cat_indices = intercepted_args["kwargs"][cat_param_name]
+
                 assert model_cat_indices == kwargs_cat_indices == expected_cat_indices
 
+                # all evals set have correct cat feature indices
+                eval_set_indices = [
+                    pool.get_cat_feature_indices()
+                    for pool in intercepted_args["kwargs"][eval_set_param_name]
+                ]
+                assert np.array([
+                    indices == model_cat_indices for indices in eval_set_indices
+                ]).all()
+
                 # catboost requires pd.DataFrame with categorical features
                 X, y = intercepted_args["args"]
                 assert isinstance(X, pd.DataFrame)
                 # all categorical features should be encoded as integers
                 for col in X[model_cat_indices].columns:
                     assert X[col].dtype == int
+
             elif model_cls == LightGBMModel:
                 assert (
                     intercepted_args["kwargs"][cat_param_name] == expected_cat_indices