Gaussian Process Software

We make software available for our research. Note that it is not 'production code', it is often just a snapshot of the software we used to produce the results in a particular paper. This makes it easier for other people to make comparisons and to reproduce our results. There are several software packages available from here, all associated with Gaussian Processes. To download these software packages you need to register, the packages are freely available for academic use, you must seek a license for commercial use.

Follow instructions on the sites to access the software.

Links to Gaussian Process Software available on line

C++ Research Software

Software	Author	Description
C++ GP-LVM	Neil D. Lawrence	GP-LVM software in C++. Currently doesn't implement the sparse algorithms, but includes dynamics and back constraints.
C++ IVM	Neil D. Lawrence	IVM Software in C++ , also includes the null category noise model for semi-supervised learning.

MATLAB Research Software

Software	Main Author	Description
Fast Gaussian Process Latent Variable Model	Neil D. Lawrence	GP-LVM using sparse approximations described at NIPS 2005 by Snelson and Ghahramani as well as extensions given by Quinonero-Candela and Rasmussen and the variational sparse approximation of Titsias.
Gaussian Process Latent Variable Model	Neil D. Lawrence	GP-LVM using the IVM based sparse approximation.
Gaussian Process Toolbox	Neil D. Lawrence	Gaussian Process Regression using sparse approximations described at NIPS 2005 by Snelson and Ghahramani as well as extensions given by Quinonero-Candela and Rasmussen and the variational approximation of Titsias.
Multiple Output Gaussian Processes	Mauricio Alvarez	Multiple output Gaussian Process Regression using sparse approximations described at NIPS 2008 by Alvarez and Lawrence as well as variational extensions.
GP Single Input Motif Toolbox	Neil D. Lawrence, Pei Gao, Antti Honkela	Gaussian Processes for Single Input Motifs using GPs for inference of transcription factor concentration in a single input motif network module.
Shared Gaussian Process Latent Variable Model	Carl Henrik Ek	GP-LVM work with Carl Henrik Ek on shared latent variable spaces.
Hierarchical Gaussian Process Latent Variable Model	Neil D. Lawrence	Hierarchical GP-LVM as presented at ICML 2007.
Gaussian Process Latent Variable Model	Neil D. Lawrence	The original GP-LVM software using sparse approximations based on the IVM.

Other Gaussian Process Software

Software	Author	Description
Bayesian Fisher's Discriminant	Tonatiuh Pena Centeno	A Gaussian process interpretation of Kernel Fisher Discriminants.
Informative Vector Machine	Neil D. Lawrence	A sparse approximation to full Gaussian processes.
Multi-task Informative Vector Machine	Neil D. Lawrence	Multi-task learning with Gaussian processes using the IVM sparse approximation
Probabilistic Point Assimilation	Nathaniel King and Neil D. Lawrence	A general fast variational method for GPs
GP Demos	Neil D. Lawrence	A set of Gaussian process demos, sampling from covariance functions etc..

This software relies on many of the toolboxes listed below.

MATLAB Toolkits

Toolbox	Description
DATASETS	Various datasets and tools for loading them.
KERN	Various utilities for computing kernels. Includes contributions by many people.
NOISE	Various noise models for Gaussian processes.
NDLUTIL	Various utilities that some toolboxes rely on.
MLTOOLS	Various Machine Learning Tools that some toolboxes rely on.
MOCAP	Tools for loading in and playing with MOCAP data.
OPTIMI	Various optimisation tools.
PRIOR	Various utilities for prior distributions.

Making Software Available

Really Reproducible Research in the Computational Sciences

I believe machine learning researchers should be making their software available at the same time they submit (or before) their papers to conference papers or journals, and I've carried out this practice since 2001. I wanted to put together the reasons why we should be doing this at some point, but it turns out that other researchers have already laid out reasons that pretty much match my own. So if you want to know why I (and why you should) make your code available that reproduces the figures in your papers please read this which was inspired by ideas of Jon Claerbout. See his white paper here. Thanks to Kevin Murphy for pointing out these papers. Neil Lawrence, 05 December 2005