RxInfer.jl is a user-friendly Julia package that automates inference in probabilistic models. It offers clean specifications of probabilistic models and inference constraints, supports streaming datasets, and allows for hybrid models with both discrete and continuous latent variables. The package is scalable, making it suitable for large models with millions of parameters and observations. It can also be extended with custom operations and supports automatic differentiation packages for parameter tuning. RxInfer leverages factor graphs for fast message passing-based probabilistic inference, outperforming sampling-based packages by several orders of magnitude. It has various applications, including tracking hidden states of dynamic systems, navigation and collision avoidance, and decision making with the Active Inference framework. Additionally, the package has community videos available for learning and reference.
https://biaslab.github.io/rxinfer-website/