Abstract
Stan is a popular declarative probabilistic programming language with ahigh-level syntax for expressing graphical models and beyond. Stan differs bynature from generative probabilistic programming languages like Church,Anglican, or Pyro. This paper presents a comprehensive compilation scheme tocompile any Stan model to a generative language and proves its correctness.This sheds a clearer light on the relative expressiveness of different kinds ofprobabilistic languages and opens the door to combining their mutual strengths.Specifically, we use our compilation scheme to build a compiler from Stan toPyro and extend Stan with support for explicit variational inference guides anddeep probabilistic models. That way, users familiar with Stan get access to newfeatures without having to learn a fundamentally new language. Overall, ourpaper clarifies the relationship between declarative and generativeprobabilistic programming languages and is a step towards making deepprobabilistic programming easier.