probability

Why we included this project

TensorFlow Probability is the standard way to add probabilistic reasoning to models built in TensorFlow, and the same code runs on JAX if you switch the backend. It ships a broad set of distributions, bijectors for building transformed and normalizing-flow models, and joint distributions that make hierarchical Bayesian models straightforward to express. Inference methods like variational Bayes and Markov chain Monte Carlo sit on automatic differentiation, so they scale to large datasets across GPUs or distributed setups. For teams that need calibrated uncertainty estimates, say in Bayesian neural networks or time-series forecasting, having these components ready to use beats implementing the sampling math from scratch.