Implementing "Vanilla" MCMC

I’ve been learning Pyro because I want to implement my own advanced MCMC algorithm as part of a research project for my studies. Only HMC and NUTS are ready “out-of-the-box” in Pyro, but I’m wondering if anyone has resources or code examples of a basic Metropolis-Hastings or Gibbs sampling algorithm.

Disclaimer: I know that these standard MCMC approaches will be functionally useless compared to NUTS or HMC, but my goal here is to get a better understanding of designing my own inference algorithm in Pyro. Any suggestions for resources regarding poutine / MCMC kernel design are appreciated.

Update: I did find this post

1 Like