Download 08-20

O Papaspiliopoulos

A note on posterior sampling from Dirichlet mixture models

Abstract: In this note we observe that the recent MCMC methods of Papaspiliopoulos & Roberts (2008) and Walker (2007) for Dirichlet mixture models are intrinsically connected and can be naturally combined to yield an algorithm which is better (in terms of mixing), faster (in terms of execution time) and easier (in terms of implementation and coding) than either of them.

Keywords: Retrospective sampling; Slice sampling; Augmentation schemes; Label switching; ; Stick-breaking priors; blocking strategies.