By this milestone, GEM should implement a multivariate linear regression model, building an adaptive Metropolis Hastings sampler from tensorflow_probability components, run, and return samples from the posterior.