Nonparametric Bayesian Biclustering

摘要: We present a probabilistic block-constant biclustering model that simultaneously clusters rows and columns of data matrix. All entries with the same row cluster column form bicluster. Each is part mixture having nonparametric Bayesian prior. The number biclusters therefore treated as nuisance para meter implicitly integrated over during simulation. Missing are completely out model, allowing us to bipass common requirement for algorithms missing values be filled before analysis, but also makes it robust h igh rates values. By using Gaussian density in bliclusters, an efficient sampling algorithm produced because bicluster parameters analytically out. several inference procedures indicat ors, including Gibbs split-merge moves. show our method competitive, if not superior, existing imputation methods, especially high rates, despite imputing co nstant entire blocks data. experiments exploratory results.