IterGP: Computation-Aware Gaussian Process Inference

This repository contains an implementation of the framework described in the paper Posterior and Computational Uncertainty in Gaussian Processes.

Installation

You can install the Python package via pip:

pip install git+https://github.com/JonathanWenger/itergp.git

Documentation and Tutorials

To understand how to use the functionality of IterGP, take a look at the API reference and the tutorials.

Datasets

Any datasets used in the experiments can be accessed via the API:

from itergp import datasets

data = datasets.uci.BikeSharing(dir="data/uci")
data.train.y
# array([ 0.20011634, -2.74432264,  0.14604912, ...,  0.40556032,
#         0.57590568, -0.54709806])

If the dataset is not already cached, it will be downloaded and cached locally.

Citation

@inproceedings{wenger2022itergp,
  author    = {Jonathan Wenger and Geoff Pleiss and Marvin Pf{\"o}rtner and Philipp Hennig and John P. Cunningham},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  keywords  = {gaussian processes, probabilistic numerics, numerical analysis},
  title     = {Posterior and Computational Uncertainty in {G}aussian processes},
  url       = {https://arxiv.org/abs/2205.15449},
  year      = {2022}
}