For testing whether Bayesian NN priors are generalization-agnostic...or nah.
Requires Python 3 & Tensorflow.
Run: python svhn_good_or_bad.py good to train on just SVHN train set, or python svhn_good_or_bad.py bad to train on the SVHN train set and a corrupted version of the test set.
Run: python svhn_is_posterior_real.py to train a BNN and also a regular neural network on the same data.