Reading a few papers to understand how to do inference with LDA. Why are Bayesian methods always so hard to implement??