An Empirical Study of Rich Subgroup Fairness for Machine Learning