These files provide simple demonstration Matlab code, corresponding to the paper:

Projecting Ising Models for Fast Mixing, Justin Domke and Xianghang Liu, Advances in Neural Information Processing Systems, 2013.

This provides a way to project Ising model parameters such that the resulting parameters will be fast mixing.  This code only provides *Euclidean* projection.

This code requires one outside file to work:

1) The minConf_TMP.m file, available as part of Mark Schmidt’s minConf package at http://www.cs.ubc.ca/~schmidtm/Software/minConf.html

Though not strictly needed, the demos show that the result is correct, by comparing to a slow, but reliable solver.  You will want to install this if you don’t believe that the result is correct.

2) The CVX optimization system, freely available from http://cvxr.com/cvx/.
