Search Algorithms and Loss Functions for Bayesian Clustering

We propose a randomized greedy search algorithm to find point estimate for random partition based on loss function and posterior Monte Carlo samples. Given the large size awkward discrete nature of space, minimization expected is challenging. Our approach stochastic series optimizations performed in order embarrassingly parallel. consider several functions, including Binder variation information. note that criticisms are result using equal penalties misclassification we show an efficient means compute with potentially unequal penalties. Furthermore, extend original information allow no increased computational costs. provide reference implementation our algorithm. Using variety examples, method produces clustering estimates better minimize obtained faster than existing methods. Supplementary materials this article available online.