test fanoval importance

Author: SimonBlanke

tests/core/custom_test_function/test_custom_test_function.py

@@ -350,3 +350,63 @@ def test_repr_contains_info(self, sphere_func):
         assert "CustomTestFunction" in repr_str
         assert "n_dim=2" in repr_str
         assert "n_evaluations=1" in repr_str
+
+
+@pytest.fixture
+def asymmetric_func_with_data():
+    """Function where x dominates: f = x^2 + 0.01*y. 100 evaluations."""
+    func = CustomTestFunction(
+        objective_fn=lambda p: p["x"] ** 2 + 0.01 * p["y"],
+        search_space={"x": (-5, 5), "y": (-5, 5)},
+    )
+    rng = np.random.default_rng(42)
+    for _ in range(100):
+        func({"x": float(rng.uniform(-5, 5)), "y": float(rng.uniform(-5, 5))})
+    return func
+
+
+class TestFanovaImportance:
+    """Test fANOVA-based parameter importance analysis."""
+
+    def test_returns_correct_structure(self, sphere_func_with_data):
+        """Returns dict with all param names, values >= 0, summing to 1."""
+        importance = sphere_func_with_data.analysis.parameter_importance(method="fanova")
+        assert set(importance.keys()) == {"x", "y"}
+        assert all(v >= 0 for v in importance.values())
+        assert sum(importance.values()) == pytest.approx(1.0, abs=1e-10)
+
+    def test_identifies_dominant_parameter(self, asymmetric_func_with_data):
+        """For f = x^2 + 0.01*y, fANOVA should assign x > 80% importance."""
+        importance = asymmetric_func_with_data.analysis.parameter_importance(method="fanova")
+        assert importance["x"] > 0.8
+
+    def test_symmetric_function_equal_importance(self):
+        """For f = x^2 + y^2, both parameters should get roughly equal weight."""
+        func = CustomTestFunction(
+            objective_fn=lambda p: p["x"] ** 2 + p["y"] ** 2,
+            search_space={"x": (-5, 5), "y": (-5, 5)},
+        )
+        rng = np.random.default_rng(42)
+        for _ in range(100):
+            func({"x": float(rng.uniform(-5, 5)), "y": float(rng.uniform(-5, 5))})
+
+        importance = func.analysis.parameter_importance(method="fanova")
+        assert abs(importance["x"] - importance["y"]) < 0.25
+
+    def test_requires_minimum_30_evaluations(self):
+        """fANOVA should raise ValueError with fewer than 30 data points."""
+        func = CustomTestFunction(
+            objective_fn=lambda p: p["x"] ** 2,
+            search_space={"x": (-5, 5)},
+        )
+        for i in range(20):
+            func({"x": float(i) - 10.0})
+
+        with pytest.raises(ValueError, match="requires at least 30"):
+            func.analysis.parameter_importance(method="fanova")
+
+    def test_outperforms_variance_on_nonlinear(self, asymmetric_func_with_data):
+        """fANOVA should capture x^2 better than linear correlation does."""
+        var_imp = asymmetric_func_with_data.analysis.parameter_importance(method="variance")
+        fanova_imp = asymmetric_func_with_data.analysis.parameter_importance(method="fanova")
+        assert fanova_imp["x"] > var_imp["x"]