Pyro

Pyro makes it possible to describe an uncertain process in Python and infer which hidden conditions could explain observed data. In its November 2017 release article, Uber introduced the system as a way to combine probabilistic modeling with deep learning.

A Pyro model mixes ordinary calculations with random choices. A second model, called a guide, learns an approximation to the hidden values' distribution given the observations. Optimizing that guide automates part of the inference work that researchers would otherwise need to implement. This separates describing a problem from building the machinery to reason about its uncertainty, while allowing neural networks inside both models and guides.