This is the official implementation of the bipartite matching experiment from the paper "Learning Randomly Perturbed Structured Predictors for Direct Loss Minimization".