Quantifying the probability of a label prediction being correct on a given test point or a given sub-population enables users to better decide how to use and when to trust machine learning derived predictors. In this work, ...