objectif

Expliquer une prédiction tabulaire d’un modèle de type gradient boosting.

code minimal

import shap
from xgboost import XGBClassifier
from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split

X, y = load_breast_cancer(return_X_y=True)
X_train, X_val, y_train, y_val = train_test_split(X, y, test_size=0.2, random_state=0, stratify=y)

model = XGBClassifier(n_estimators=200, max_depth=3, learning_rate=0.1, subsample=0.9, colsample_bytree=0.9, tree_method="hist", random_state=0)
model.fit(X_train, y_train)

explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X_val[:10])
print(len(shap_values) == 10)

utilisation

# Valeur de base (expected value)
base = explainer.expected_value
print(base is not None)

variante(s) utile(s)

# Force plot / waterfall (les figures sont objets Matplotlib)
# shap.plots.waterfall(shap.Explanation(values=shap_values[0], base_values=base, data=X_val[0]))
print(True)

notes

• SHAP nécessite des dépendances graphiques pour afficher; pour CPU-only, TreeExplainer est rapide sur modèles d’arbres.