No Pseudolikelihood for Me: Training Potts Models with Contrastive Divergence for Protein Design - View it on GitHub