mini-gpr

hyperparameter-optimised GPR models trained
on a toy 1D dataset of increasing size

mini-gpr is a minimal reference implementation of Gaussian Process Regression in pure NumPy, made primarily for my own learning.

Features of mini-gpr include:

• implementations of a full (GPR) and low rank (SoR) GPR models

• implementations of several common kernels

• Model and Kernel base classes for easy extension

• automated hyperparameter optimisation against a range of objectives via mini_gpr.opt.optimise_model

• strong typing using the jaxtyping library

This documentation includes:

• a tutorial in the form of a collection of Jupyter notebooks

• a complete API reference for this small, stand-alone package