Bayes-Newton—A Gaussian process library in JAX, with a unifying view of approximate Bayesian inference as variants of Newton's method.
#21
Bug fix: #20
Small bug fix when flattening during Gauss-Newton Hessian computation.
Updates to allow compatibility with latest JAX and Objax versions. Bug fix in partial Gauss Newton.
The sparse PEP energy is now correct, and the stochastic sparse PEP algorithm is now fully implemented.
A first-order variational inference method has been added as a baseline for experiments.
Some additional documentation has been added, and some broken tests have been fixed.
Remove unnecessary files
A major refactor including all the previously private development code.
Many inference methods and models implemented. Some test coverage.