Paper extending Stan language with variational inference guides and translating models to Pyro

Really don’t know whether this has any practical relevance, but interesting paper anyway as they show how to add guides for non-black-box variational inference using guides and they translate models to Pyro and make some performance comparisons

Extending Stan for Deep Probabilistic Programming


If they can do Stan --> Pyro could they also do Pyro --> Stan?

From what I understand, they used our AST with a new code generator they wrote into the C++. We’re trying to make this easier with the new language design/architecture that @seantalts is designing.