Recommendation of data-free class-incremental learning algorithms by simulating future data