I might be wrong here, but you also need uncertainty estimate for training/test data set in relation to empirical domain, at least in cases with prohibitively l

I might be wrong here, but you also need uncertainty estimate for training/test data set in relation to empirical domain, at least in cases with prohibitively large problem state space.