Deep universal probabilistic programming with Python and PyTorch

Getting Started | Documentation | Contributing

Pyro is a flexible, scalable deep probabilistic programming library built on PyTorch. Notably, it was designed with these principles in mind:

• Universal: Pyro is a universal PPL -- it can represent any computable probability distribution.
• Scalable: Pyro scales to large data sets with little overhead compare to hand-written code.
• Minimal: Pyro is agile and maintainable. It is implemented with a small core of powerful, composable abstractions.
• Flexible: Pyro aims for automation when you want it, control when you need it. This is accomplished through high-level abstractions to express generative and inference models, while allowing experts easy-access to customize inference.

Pyro is in an alpha release. It is developed and used by Uber AI Labs.

Installation

First install PyTorch.

Most features of Pyro work on PyTorch's 0.2 release, but some features are only available on PyTorch's master branch (e.g. pyro.SVI(... enum_discrete=True and pyro.SVI(..., num_particles=100) require PyTorch more recent than 0.2). To use these features, we recommend installing PyTorch from source. We have verified that commit f964105 supports all of Pyro. To get this version, run git clone https://github.com/pytorch/pytorch.git && git checkout f964105.

Install via pip:

Python 2.7.*:

pip install pyro-ppl

Python 3.5:

pip3 install pyro-ppl

Install from source:

git clone git@github.com:uber/pyro.git
cd pyro
pip install .