Comment on Bayesian Cluster Analysis: Point Estimation and Credible Balls by Wade and Ghahramani

This article discusses the Wade and Ghahramani’s (2018) paper on a new estimator for clustering structures based on the variation of information (VI) metric. The present discussion focuses on the estimation of concentration parameter of the Dirichlet process. In estimating the clustering structure, the concentration parameter is integrated out and the marginal posterior distribution of the random partition is used to evaluate the posterior loss. Here we propose to use the optimal VI for model selection.