explainer = shap.Explainer(self.model) sv = np.array(explainer.shap_values(self.X_valid)) ev = np.array(explainer.expected_value) idx = 0 print(self._feature_cols) # 921 feature names print(self.X_valid.shape) # (57353, 921) shap.plots.waterfall(shap.Explanation(sv[1][idx], ev[idx], data=self.X_valid.iloc[idx], feature_names=self._feature_cols), max_display=20) Even after setting max_display to 20, I still only ever get the same three features, no idea why. The three features aren't even the top in self.model.feature_importances_, so I have no idea why these in particular are being displayed, but not the hundreds of other features. I tried both Explainer and TreeExplainer, For reference, my model is defined as: self.model = lgb.LGBMClassifier( device='gpu', boosting_type='gbdt', max_depth=6, n_estimators=500, learning_rate=0.05, objective='multiclass', num_class=len(self.fe.bins_labels), metric='multi_logloss', class_weight=class_weight_dict, random_state=Trainer.seed, min_child_samples=60, min_split_gain=0.1, verbose=-1 ) Continue reading...
