[BUG] - bugfix, GridSearchCV scoring=None defaults to CRPS

skpro/model_selection/_tuning.py

@@ -8,6 +8,7 @@
 import pandas as pd
 from sklearn.model_selection import ParameterGrid, ParameterSampler, check_cv
 
+from skpro.metrics import CRPS
 from skpro.benchmarking.evaluate import evaluate
 from skpro.regression.base._delegate import _DelegatedProbaRegressor
 from skpro.utils.parallel import parallelize
@@ -108,6 +109,8 @@ def _fit(self, X, y, C=None):
 
         # scoring = check_scoring(self.scoring, obj=self)
         scoring = self.scoring
+        if scoring is None:
+            scoring = CRPS()
         scoring_name = f"test_{scoring.name}"
 
         backend = self.backend

skpro/model_selection/tests/__init__.py

@@ -0,0 +1 @@
+"""Tests for model selection utilities."""

skpro/model_selection/tests/test_tuning.py

@@ -0,0 +1,53 @@
+"""Tests for model selection tuning utilities."""
+
+import pandas as pd
+import pytest
+from sklearn.model_selection import KFold
+
+from skpro.model_selection import GridSearchCV, RandomizedSearchCV
+from skpro.regression.dummy import DummyProbaRegressor
+from skpro.tests.test_switch import run_test_module_changed
+
+
+def _get_test_data():
+    X = pd.DataFrame({"x": [1, 2, 3, 4]})
+    y = pd.DataFrame({"y": [1.0, 2.0, 3.0, 4.0]})
+    return X, y
+
+
+@pytest.mark.skipif(
+    not run_test_module_changed("skpro.model_selection"),
+    reason="Test only if skpro.model_selection has been changed",
+)
+def test_gridsearch_scoring_none_defaults_to_crps():
+    """GridSearchCV should use CRPS when scoring is None."""
+    X, y = _get_test_data()
+
+    gscv = GridSearchCV(
+        estimator=DummyProbaRegressor(),
+        cv=KFold(n_splits=2),
+        param_grid={"strategy": ["empirical"]},
+        scoring=None,
+    )
+    gscv.fit(X, y)
+
+    assert "mean_test_CRPS" in gscv.cv_results_.columns
+
+
+def test_randomizedsearch_scoring_none_defaults_to_crps():
+    """RandomizedSearchCV should use CRPS when scoring is None."""
+    X, y = _get_test_data()
+
+    rscv = RandomizedSearchCV(
+        estimator=DummyProbaRegressor(),
+        cv=KFold(n_splits=2),
+        param_distributions={"strategy": ["empirical"]},
+        n_iter=1,
+        scoring=None,
+        backend_params={"n_jobs": 1},
+    )
+    rscv.fit(X, y)
+
+    assert "mean_test_CRPS" in rscv.cv_results_.columns
+
+