Welcome to our anonymous NIPS submission. All code remains copyright of the anonymous authors, released under the GPL license (see LICENSE)

There are two directories in this repo. mcmcGP contains a python module which implements the method. experiments contains subdirectories with code to replicate the examples in the paper.

To work with our code, you'll need to instal our little module. We recommend running python setup.py build_ext --inplace to build the cython modeules, and then add mcmcGP to your PYTHONPATH.

The code needs numpy, scipy, matplotlib, pandas, jug (https://github.com/luispedro/jug) and GPy (https://github.com/SheffieldML/GPy). The standard version of jug will suffice (pip install jug), but the bleeding edge of GPy is needed (see their install instructions).

Running the code takes a long time. There are hundreds of models to fit! you'll need to use jug for image_demo, spatial_demo, etc.

A good place to start is simple_classification, which should run in about a minute without jug. We recommend IPython for interacting with the code and figures.