Large Scale Gaussian Processes using Grid-Structured Eigenfunction (GRIEF) kernels.
Only python 2 is supported.
An up-to-date Anaconda distribution is recommended in addition to the following non-standard dependencies (using pip):
pip install GPy==1.6.1 # v1.6.1 tested but other versions should also work
pip install tabulate # this is a lightweight package used for printing
Also required is the forked version of the py-mcmc library:
git clone https://github.com/treforevans/py-mcmc.git
cd py-mcmc
python setup.py install
- Type-II inference with GRIEF kernels: This example demonstrates the type-II training and inference procedure with GP-GRIEF.
- Type-I inference with GRIEF kernels: This example demonstrates how to perform fully Bayesian type-I inference with a runtime that is independent of the training dataset size.
The underlying algorithms are based on the following 2018 ICML paper (found here):
@InProceedings{evans_gp-grief,
title={Scalable {G}aussian Processes with Grid-Structured Eigenfunctions ({GP}-{GRIEF})},
author={Evans, Trefor W and Nair, Prasanth B},
booktitle={International Conference on Machine Learning},
pages={1416--1425},
year={2018}
}