integrates intercept of xgb models directly into values of the TreeModel

shapiq/explainer/tree/conversion/xgboost.py

@@ -96,6 +96,7 @@ def _convert_xgboost_tree_as_df(
 
     # pandas can't chill https://stackoverflow.com/q/77900971
     with pd.option_context("future.no_silent_downcasting", True):
+        values = tree_df["Gain"].values * scaling + intercept  # add intercept to all values
         tree_model = TreeModel(
             children_left=tree_df["Yes"]
             .replace(convert_node_str_to_int)
@@ -111,9 +112,9 @@ def _convert_xgboost_tree_as_df(
             .values,
             features=tree_df["Feature"].values,
             thresholds=tree_df["Split"].values,
-            values=tree_df["Gain"].values * scaling,  # values in non-leaf nodes are not used
+            values=values,  # values in non-leaf nodes are not used
             node_sample_weight=tree_df["Cover"].values,
-            empty_prediction=intercept,
+            empty_prediction=None,
             original_output_type=output_type,
         )
 

shapiq/explainer/tree/explainer.py

@@ -14,31 +14,6 @@
 from .validation import validate_tree_model
 
 
-def set_baseline_value(model: Any, treeshapiq_explainers: list[TreeSHAPIQ]) -> float:
-    """Sets the baseline value for the interaction values.
-
-    Tries to set the baseline value for the interaction values from a model.
-
-    Args:
-        model: The model to explain.
-        treeshapiq_explainers: The treeSHAP-IQ explainers.
-
-    Returns:
-        The baseline value for the interaction values.
-    """
-    # default value for the baseline provided by ensembles
-    # works for sklearn decision trees and random forests
-    baseline_value = sum([treeshapiq.empty_prediction for treeshapiq in treeshapiq_explainers])
-    try:  # xgboost models have base_score/intercept_
-        base_score = model.base_score
-        if base_score is None:
-            base_score = float(model.intercept_[0])
-        baseline_value = base_score if base_score is not None else baseline_value
-    except AttributeError:
-        pass
-    return baseline_value
-
-
 class TreeExplainer(Explainer):
     """
     The explainer for tree-based models using the TreeSHAP-IQ algorithm.
@@ -96,8 +71,7 @@ def __init__(
         self._treeshapiq_explainers: list[TreeSHAPIQ] = [
             TreeSHAPIQ(model=_tree, max_order=self._max_order, index=index) for _tree in self._trees
         ]
-
-        self.baseline_value = set_baseline_value(self.model, self._treeshapiq_explainers)
+        self.baseline_value = self._compute_baseline_value()
 
     def explain(self, x: np.ndarray) -> InteractionValues:
         # run treeshapiq for all trees
@@ -112,3 +86,17 @@ def explain(self, x: np.ndarray) -> InteractionValues:
             for i in range(1, len(interaction_values)):
                 final_explanation += interaction_values[i]
         return final_explanation
+
+    def _compute_baseline_value(self) -> float:
+        """Computes the baseline value for the explainer.
+
+        The baseline value is the sum of the empty predictions of all trees in the ensemble.
+
+        Returns:
+            The baseline value for the explainer.
+        """
+
+        baseline_value = sum(
+            [treeshapiq.empty_prediction for treeshapiq in self._treeshapiq_explainers]
+        )
+        return baseline_value

tests/tests_explainer/tests_tree_explainer/test_tree_explainer.py

@@ -135,10 +135,8 @@ def test_xgboost_reg(xgb_reg_model, background_reg_data):
     # explainer_shap = shap.TreeExplainer(model=xgb_reg_model)
     # x_explain_shap = background_reg_data[explanation_instance].reshape(1, -1)
     # sv_shap = explainer_shap.shap_values(x_explain_shap)[0]
-    # baseline_shap = explainer_shap.expected_value
     sv_shap = [-2.555832, 28.50987, 1.7708225, -7.8653603, 10.7955885, -0.1877861, 4.549199]
     sv_shap = np.asarray(sv_shap)
-    baseline_shap = -2.5668228
 
     # compute with shapiq
     explainer_shapiq = TreeExplainer(model=xgb_reg_model, max_order=1, index="SV")
@@ -147,12 +145,11 @@ def test_xgboost_reg(xgb_reg_model, background_reg_data):
     sv_shapiq_values = sv_shapiq.get_n_order_values(1)
     baseline_shapiq = sv_shapiq.baseline_value
 
-    assert baseline_shap == pytest.approx(baseline_shapiq, rel=1e-4)
-    assert np.allclose(sv_shap, sv_shapiq_values, rtol=1e-4)
+    assert np.allclose(sv_shap, sv_shapiq_values, rtol=1e-5)
 
     # get prediction of the model
     prediction = xgb_reg_model.predict(x_explain_shapiq.reshape(1, -1))
-    assert prediction == pytest.approx(baseline_shapiq + np.sum(sv_shapiq_values), rel=1e-2)
+    assert prediction == pytest.approx(baseline_shapiq + np.sum(sv_shapiq_values), rel=1e-5)
 
 
 def test_xgboost_clf(xgb_clf_model, background_clf_data):
@@ -173,7 +170,6 @@ def test_xgboost_clf(xgb_clf_model, background_clf_data):
     # print(sv_shap)
     sv = [-0.00545454, -0.15837783, -0.17675081, -0.24213657, 0.00247543, 0.00988865, -0.01564346]
     sv_shap = np.array(sv)
-    baseline_shap = 0.5
 
     # compute with shapiq
     explainer_shapiq = TreeExplainer(
@@ -184,14 +180,14 @@ def test_xgboost_clf(xgb_clf_model, background_clf_data):
     sv_shapiq_values = sv_shapiq.get_n_order_values(1)
     baseline_shapiq = sv_shapiq.baseline_value
 
-    assert baseline_shap == pytest.approx(baseline_shapiq, rel=1e-4)
-    assert np.allclose(sv_shap, sv_shapiq_values, rtol=1e-4)
+    # assert baseline_shap == pytest.approx(baseline_shapiq, rel=1e-4)
+    assert np.allclose(sv_shap, sv_shapiq_values, rtol=1e-5)
 
     # get prediction of the model (as the log odds)
     prediction = xgb_clf_model.predict(x_explain_shapiq.reshape(1, -1), output_margin=True)[0][
         class_label
     ]
-    assert prediction == pytest.approx(baseline_shapiq + np.sum(sv_shapiq_values), rel=2e-2)
+    assert prediction == pytest.approx(baseline_shapiq + np.sum(sv_shapiq_values), rel=1e-5)
 
 
 def test_xgboost_shap_error(xgb_clf_model, background_clf_data):
@@ -220,7 +216,6 @@ def test_xgboost_shap_error(xgb_clf_model, background_clf_data):
     # print(baseline_shap)
     sv = [-0.00163636, 0.05099502, -0.13182959, -0.44538185, 0.00428653, -0.04872373, -0.01370917]
     sv_shap = np.array(sv)
-    baseline_shap = 0.5
 
     # setup shapiq TreeSHAP
     explainer_shapiq = TreeExplainer(
@@ -230,9 +225,8 @@ def test_xgboost_shap_error(xgb_clf_model, background_clf_data):
     sv_shapiq = explainer_shapiq.explain(x=x_explain_shapiq)
     sv_shapiq_values = sv_shapiq.get_n_order_values(1)
 
-    # the baseline scores should be the same as with SHAP but the values should be different
-    assert baseline_shap == pytest.approx(sv_shapiq.baseline_value, rel=1e-4)
-    assert not np.allclose(sv_shap, sv_shapiq_values, rtol=1e-4)
+    # the SHAP sv values should be different from the shapiq values
+    assert not np.allclose(sv_shap, sv_shapiq_values, rtol=1e-5)
 
     # when we round the model thresholds of the xgb model (thresholds decide weather a feature is
     # used or not) -> then suddenly the shap and shapiq values are the same, which points to the
@@ -247,4 +241,4 @@ def test_xgboost_shap_error(xgb_clf_model, background_clf_data):
     sv_shapiq_rounded_values = sv_shapiq_rounded.get_n_order_values(1)
 
     # now the values surprisingly are the same
-    assert np.allclose(sv_shap, sv_shapiq_rounded_values, rtol=1e-4)
+    assert np.allclose(sv_shap, sv_shapiq_rounded_values, rtol=1e-5)