On Bayesian analysis of mixtures with an unknown number of components-英文文献.pdf

On Bayesian analysis of mixtures with an unknown numb er of comp onents  y z Sylvia Richardson Peter J Green INSERM France University of Bristol UK Revised version Octob er Summary New metho dology for fully Bayesian mixture analysis is develop ed making use of reversible jump Markov chain Monte Carlo metho ds that are capable of jumping b etween the parameter subspaces corresp onding to dierent numb ers of comp onents in the mixture A sample from the full join t distribution of all unknown variables is thereby generated and this can b e used as a basis for a thorough presentation of many asp ects of the p osterior distribution The metho dology is applied here to the analysis of univariate normal mixtures using a hierarchical prior mo del that oers an approach to dealing with weak prior information while avoiding the mathematical pitfalls of using improp er priors in the mixture context Some key words Birth and death pro cess Classication Galaxy data Heterogeneity Lake acidity data Markov chain Monte Carlo Normal mixtures Predictive distribution Reversible jump algorithms Sensitivity analysis Intro duction This article is a contribution to the metho dology of fully Bayesian mixture mo delling We stress the word fully in two senses First we mo del the numb er of comp onents and the mixture comp onent parameters join tly and base inference ab out these quantities on their p osterior probabilities This is in contrast to most previous Bayesian treatments of mixture estimation whi