pyGPs - A Package for Gaussian Processes

About the package

pyGPs is a library hosting Python implementations of Gaussian processes (GPs) for machine learning. pyGPs bridges the gap between systems designed primarily for users, who mainly want to apply GPs and need basic machine learning routines for model training, evaluation, and visualiztion, and expressive systems for developers, who focus on extending the core functionalities as covariance and likelihood functions, as well as inference techniques.

The software package is released under the BSD 2-Clause (FreeBSD) License. Copyright (c) by Marion Neumann, Shan Huang, Daniel Marthaler, & Kristian Kersting, Feb.2014

Further, it includes implementations of

  • implemented in python by Roland Memisevic 2008, following minimize.m (Copyright (c) Carl Edward Rasmussen (1999-2006))
  • (Copyright (c) Ian T Nabney (1996-2001))
  • (Copyright (c) Hannes Nickisch 2010-01-10)
  • FITC functionality (following matlab implementations under Copyright (c) by Ed Snelson, Carl Edward Rasmussen and Hannes Nickisch, 2011-11-02)

The most recent stable release is pyGPs v1.3.1. See changelog for a list of changes to the previous release. If you observe problems or bugs, please contact us. You can download the developer’s guide and manual containing an API here: API and manual. You can also download a procedual implementation of GP functionality from Github. However, the procedual version will not be supported in future.

  • Marion Neumann [marion dot neumann at uni-bonn dot de]
  • Shan Huang [shan dot huang at iais dot fraunhofer dot de]
  • Daniel Marthaler [dan dot marthaler at gmail dot com]
  • Kristian Kersting [kristian dot kersting at cs dot tu-dortmund dot de]

The following persons helped to improve this software: Roman Garnett, Maciej Kurek, Hannes Nickisch, Zhao Xu, and Alejandro Molina.

This work is partly supported by the Fraunhofer ATTRACT fellowship STREAM.