cornellius-gp/gpytorch
A Gaussian process library implemented in PyTorch that provides scalable, GPU-accelerated inference through linear operator techniques.
Velocity · 7d
+1.2
★ / day
Trend
→steady
star history
GPyTorch implements Gaussian processes using PyTorch, distinguishing itself from traditional GP approaches by leveraging numerical linear algebra techniques like preconditioned conjugate gradients instead of Cholesky decomposition. This design enables better GPU utilization and scalability. The library provides state-of-the-art implementations of scalable GP methods including SKI/KISS-GP, stochastic Lanczos expansions, LOVE, and SKIP for handling large datasets efficiently.