eb73ec72ed6c28b774a4d4220a9a02bac2f8be81,bugbug/model.py,Model,train,#Model#,32

Before Change

        if len(feature_names):
            print("Feature ranking (top 20 features):")
            indices = np.argsort(self.clf.feature_importances_)[::-1][:20]
            for i in range(len(indices)):
                print("{}. \"{}\" ({})".format(i + 1, feature_names[indices[i]], self.clf.feature_importances_[indices[i]]))

        y_pred = self.clf.predict(X_test)

        print("No confidence threshold - {} classified".format(len(y_test)))
        print(metrics.confusion_matrix(y_test, y_pred, labels=class_names))

After Change

            // Calculate the values that represent the fraction of the model output variability attributable
            // to each feature across the whole dataset.
            shap_sums = np.abs(shap_values).sum(0)
            rel_shap_sums = shap_sums / shap_sums.sum()
            indices = np.argsort(rel_shap_sums)[::-1][:20]
            for i, index in enumerate(indices):
                print("{}. \"{}\" ({})".format(i + 1, feature_names[index], rel_shap_sums[index]))

        y_pred = self.clf.predict(X_test)

        print("No confidence threshold - {} classified".format(len(y_test)))
        print(metrics.confusion_matrix(y_test, y_pred, labels=class_names))

Instances