A New Method for Tracking Configuration for Dirichlet Process Sampling

The method of fitting a hierarchical model with Dirichlet process mixing is a versatile tool for data analysts. It has been applied to density estimation, classification, clustering, and high dimensional data analysis. Many computing algorithms have been proposed to evaluate this mixture. Different labels in the algorithm that assign data points into clusters may actually yield the same partition configuration. This paper makes this notion rigorous by establishing an equivalence theorem. Thus, we would recommend adding the step of checking for equivalent configurations to the algorithms for evaluating hierarchical Dirichlet process mixing models for improved results, especially when cluster assignments are the major goals of the analysis.

DOI: http://dx.doi.org/10.4038/sljastats.v5i4.7781