Fast and consistent algorithm for the latent block model

The latent block model is used to simultaneously rank the rows and columns of a matrix reveal structure. algorithms for estimation are often time consuming. However, recent work shows that log-likelihood ratios equivalent under complete observed (with unknown labels) models groups posterior distribution converge as size data increases Dirac mass located at actual configuration. Based on these observations, algorithm Largest Gaps proposed in this paper perform clustering using only marginals matrix, when number blocks very small with respect whole case binary data. In addition, selection method incorporated proof its consistency. Thus, studying simplistic configurations (few compared or contrasting blocks) complex useless since already give good parameter classification estimates.