This is an implementation of FOPPL, an S-expression based probabilistic
programming language described in [1]. See the resources/examples directory
for a list of FOPPL programs.
- Compiles FOPPL programs to a graphical model representation.
- Able to perform automatic differentiation of simple, first-order functions.
- Inference algorithms: Metropolis within Gibbs and HMC.
- Also supports inference of higher-order models using an evaluation-based interpreter.
- Supports the PPX protocol. This means this can be used as an inference engine for models written in a language without probabilistic constructs.
- Clojure 1.8+
- Anglican 1.0+
$ lein run [foppl-src]
[1] J. W. van de Meent, B. Paige, H. Yang, and F. Wood, “Introduction to Probabilistic Programming,” Foundations and Trends in Machine Learning, pp. in review, 2018. arxiv.org