marcosfelt / pyrff

pyrff: Python implementation of random fourier feature approximations for gaussian processes

pyrff: Approximating Gaussian Process samples with Random Fourier Features

This project is a Python implementation of random fourier feature (RFF) approximations [1].

It is heavily inspired by the implementations from [2, 3] and generalizes the implementation to work with GP hyperparameters obtained from any GP library.

Examples are given as Jupyter notebooks for GPs fitted with PyMC3 and scikit-learn:

• Example_RFF_PyMC3_1D
• Example_RFF_PyMC3_2D
• Thompson sampling, 1D with sklearn

Installation

pyrff is released on PyPI:

pip install pyrff

Usage and Citing

pyrff is licensed under the GNU Affero General Public License v3.0.

Head over to Zenodo to generate a BibTeX citation for the latest release.

References

1. Hernández-Lobato, 2014 paper, code
2. PES implementation in Cornell-MOE code
3. Bradford, 2018 paper, code