An implementation of the conjugate gradient lower bound method described in Tighter bounds on the log Marginal likelihood of Gaussian Process regression. This method can be used for scalable hyperparameter selection for Gaussian process regression models with Gaussian likelihood using approximate Empirical Bayes'.
The repo has two version of CGLB model:
- TensorFlow: you need to install only GPflow and its dependencies.
- PyTorch: you should install GPytorch and its dependencies.
The command line interface is based on click, and with xpert experiment manager you can run and organize many experiments on different GPUs (or CPU).
You can find the full list of requirements at requirements.txt.
Install (develop):
$ pip install -r requirements.txt
$ pip install -e .
with CLI:
$ python cli.py --keops -b torch -l "./logs" -s 0 -t fp64 train -n 2000 -d snelson1d cglb -k Matern32 -m cglb -i ConditionalVariance -M 1024
with xpert:
$ xpert xpert-main.toml
@inproceedings{artemevburt21tighter,
title = {Tighter Bounds on the Log Marginal Likelihood of Gaussian Process Regression Using Conjugate Gradients},
author = {Artemev, Artem and Burt, David R. and van der Wilk, Mark},
booktitle = {Proceedings of the 38th International Conference on Machine Learning},
pages = {362--372},
year = {2021},
}