pyro

Pyro is a deep probabilistic programming library built on PyTorch. According to its README, it is designed around four principles: it is universal, meaning it can represent any computable probability distribution; scalable, with little overhead compared to hand-written code; minimal, implemented with a small core of composable abstractions; and flexible, offering high-level abstractions for generative and inference models while allowing experts to customize inference. Originally developed at Uber AI, Pyro is now maintained by community contributors, including a team at the Broad Institute, and became a Linux Foundation project in 2019. The project provides getting-started examples, documentation, a community forum, and contribution guidelines. Installation options include installing the stable release via pip (`pip install pyro-ppl`), installing from source, installing with extra packages for the examples and tutorials (`pip install pyro-ppl[extras]`), or installing the development branch directly from GitHub. Docker container instructions are also referenced. The README includes a citation for the JMLR paper describing Pyro.