Interpretability and performance: Can the same model achieve both?