jejjohnson / GPfY

Gaussian processes with spherical harmonic features in JAX

$GP \mathcal{f} Y_\ell^m$

A lightweight library in JAX for Gaussian process with spherical kernels and sparse spherical harmonic inducing features.

$GP \mathcal{f} Y_\ell^m$ is based on the simple flax.struct dataclass. It implements (Eleftheriadis et al. 2023), which revisits the Sparse Gaussian Process with Spherical Harmonic features from Dutordoir et al. 2020, and introduces:

1. PolynomialDecay kernel with "continuous" depth.
2. Sparse orthogonal basis derived from SphericalHarmonics features with phase truncation.

Installation

Latest (stable) release from PyPI

pip install gpfy

Development version

Alternatively, you can install the latest GitHub develop version. First create a virtual enviroment via conda:

conda create -n gpfy_env python=3.10.0
conda activate gpfy_env

Then clone a copy of the repository to your local machine and run the setup configuration in development mode:

git clone https://github.com/stefanosele/GPfY.git
cd GPfY
make install

This will automatically install all required dependencies.

Finally you can check the installation via running the tests:

make test