fix: Resolve LightGBM eval_set deprecation and NumPy/joblib compatibility issues

packages/openstef-beam/src/openstef_beam/evaluation/metric_providers.py

@@ -227,8 +227,8 @@ def compute_deterministic(
         )
 
         metrics: MetricsDict = {}
-        metrics["num_predicted_peaks"] = cm.true_positives.sum() + cm.false_positives.sum()
-        metrics["num_true_peaks"] = cm.true_positives.sum() + cm.false_negatives.sum()
+        metrics["num_predicted_peaks"] = int(cm.true_positives.sum() + cm.false_positives.sum())
+        metrics["num_true_peaks"] = int(cm.true_positives.sum() + cm.false_negatives.sum())
         peak_pr = precision_recall(cm, effective=False)
         effective_pr = precision_recall(cm, effective=True)
         metrics["precision"], metrics["recall"] = peak_pr

packages/openstef-models/src/openstef_models/models/component_splitting/linear_component_splitter.py

@@ -87,7 +87,7 @@ class LinearComponentSplitter(ComponentSplitter):
         ...     components=[EnergyComponentType.SOLAR, EnergyComponentType.WIND, EnergyComponentType.OTHER],
         ... )
         >>> splitter = LinearComponentSplitter(config)
-        >>> components = splitter.predict(time_series_data) # doctest: +SKIP
+        >>> components = splitter.predict(time_series_data)  # doctest: +SKIP
     """
 
     _config: LinearComponentSplitterConfig

packages/openstef-models/src/openstef_models/models/forecasting/gblinear_forecaster.py

@@ -201,7 +201,7 @@ def model_post_init(self, _context: object, /) -> None:
             # Boosting structure control
             feature_selector=self.hyperparams.feature_selector,
             updater=self.hyperparams.updater,
-            quantile_alpha=[float(q) for q in self.quantiles],
+            quantile_alpha=sorted([float(q) for q in self.quantiles]),
             top_k=self.hyperparams.top_k if self.hyperparams.feature_selector == "thrifty" else None,
             # Objective
             objective=get_objective_function(function_type=self.hyperparams.objective, quantiles=self.quantiles)

packages/openstef-models/src/openstef_models/utils/multi_quantile_regressor.py

@@ -6,6 +6,7 @@
 Designed to work with scikit-learn compatible regressors that support quantile regression.
 """
 
+import inspect
 import logging
 
 import numpy as np
@@ -89,15 +90,27 @@ def fit(
 
         for model in self._models:
             # Check if early stopping is supported
-            # Check that eval_set is supported
             if eval_set is None and "early_stopping_rounds" in self.hyperparams:
                 model.set_params(early_stopping_rounds=None)
 
             if eval_set is not None and self.learner_eval_sample_weight_param is not None:
-                kwargs["eval_set"] = [
-                    (x_array if eval_x is X else np.asarray(eval_x), eval_y) for eval_x, eval_y in eval_set
-                ]
-                kwargs[self.learner_eval_sample_weight_param] = eval_sample_weight
+                fit_signature = inspect.signature(getattr(model, "fit"))  # noqa: B009
+                has_eval_x_param = "eval_X" in fit_signature.parameters
+
+                if has_eval_x_param:
+                    # Extract X and y from eval_set tuples for LightGBM
+                    eval_x_data = tuple((x_array if eval_x is X else np.asarray(eval_x)) for eval_x, _ in eval_set)
+                    eval_y_data = tuple(eval_y for _, eval_y in eval_set)
+                    kwargs["eval_X"] = eval_x_data
+                    kwargs["eval_y"] = eval_y_data
+                    kwargs["eval_sample_weight"] = eval_sample_weight
+                else:
+                    # XGBoost uses eval_set
+                    kwargs["eval_set"] = [
+                        (x_array if eval_x is X else np.asarray(eval_x), eval_y) for eval_x, eval_y in eval_set
+                    ]
+                    kwargs[self.learner_eval_sample_weight_param] = eval_sample_weight
+
                 if "early_stopping_rounds" in self.hyperparams:
                     model.set_params(early_stopping_rounds=self.hyperparams["early_stopping_rounds"])
 

packages/openstef-models/tests/unit/transforms/time_domain/test_holiday_features_adder.py

@@ -27,8 +27,8 @@
     (date(2025, 4, 26), "King's Day", "king_s_day"),
     (date(2025, 5, 5), "Liberation Day", "liberation_day"),
     (date(2025, 5, 29), "Ascension Day", "ascension_day"),
-    (date(2025, 6, 8), "Whit Sunday", "whit_sunday"),
-    (date(2025, 6, 9), "Whit Monday", "whit_monday"),
+    (date(2025, 6, 8), "Pentecost", "pentecost"),
+    (date(2025, 6, 9), "Pentecost Monday", "pentecost_monday"),
     (date(2025, 12, 25), "Christmas Day", "christmas_day"),
     (date(2025, 12, 26), "Second Day of Christmas", "second_day_of_christmas"),
 ]

pyproject.toml

@@ -192,6 +192,9 @@ ini_options.filterwarnings = [
   # that Python cannot propagate. Scope the ignore to CPython's fixed ctypes-callback
   # wording so genuine unraisable exceptions still fail.
   "ignore:.*Exception ignored on calling ctypes callback function.*:pytest.PytestUnraisableExceptionWarning",
+  # joblib upstream: NumPy 2.5+ deprecated array.shape assignment. joblib 1.5.3 uses this
+  # in its unpickling code and cannot be fixed without upgrading joblib (not yet available).
+  "ignore:.*Setting the shape on a NumPy array has been deprecated.*:DeprecationWarning:joblib",
   # Ignore fork() deprecation warning from multiprocessing on macOS
   # This warning occurs when pytest (multi-threaded) spawns processes using fork()
   # Using spawn instead breaks test infrastructure (fixtures not picklable)