RxInfer.Jl

RxInfer.jl is a Julia package for Bayesian inference. It combines message passing on factor graphs with automatic differentiation, so that probabilistic models can be specified in familiar mathematical notation and solved efficiently without hand deriving the inference procedure. It is aimed at researchers and engineers working on machine learning, signal processing and statistical modeling in Julia. Users write a generative model with a model macro, add optional variational constraints such as a mean field factorization, and call an infer function on their data. The site lists reactive inference for streaming datasets, support for hybrid models that mix discrete and continuous latent variables, scalability to models with millions of parameters and observations, and an extensible design that accepts custom operations. Active inference is also featured on the home page. The site provides documentation, a set of worked examples and interactive diagrams that explain how nodes exchange messages in a factor graph. A community of developers and researchers is referenced for support. The vendor publishes no pricing. The package is presented as an open source library, and the site does not list any paid plans or hosted service.