Towards and Efficient Algorithm for Computing the Reduced Mutual Information

In [1], Newman et al. introduced the Reduced Mutual Information (RMI), a measure of similarity between two partitions set useful in clustering and community detection. The computation RMI requires counting amount contingency tables with fixed row column sums, #P-complete problem, for which authors suggest to use analytical approximations that work general, but other not so pathological cases these give highly inaccurate approximations. We propose hybrid scheme based on combining existing Markov chain Monte Carlo methods make more accurate estimates number all cases.