What is this repository for?
This is a Gaussian Process multi-class classification toolbox, in which Laplace Approximation is used for inference and maximising marginal likelihood is adapted to optimise the hyper-parameters of kernel functions.
Who need this toolbox?
- This toolbox is design for those who want to solve multi-class classification and require the full predictive probabilities.
How do I get set up?
(1) As this toolbox supports all kernels provided by GPML, you need to add GPML's toolbox (http://www.gaussianprocess.org/gpml/code/matlab/doc/) to the path.
(2) run startup.m (GPML) to setup environment for GPML's toolbox.
(3) run demo.m (multi-class GPC)
This toolbox is mainly following GPML (Book: Gaussian Process for Machine Learning). We implement GP multi-class classification because it is not provided by GPML's toolbox.