Evaluating the inverse decision-making approach to preference learning