Cahiers du CEREMADE

Unité Mixte de Recherche du C.N.R.S. N°7534
 
Abstract : We consider several perfect samplers for posterior distributions associated mixtures of exponential families, including a perfect slice sampler in the spirit of Mira (1999). The methods rely on a marginalisation akin to Rao-Blackwellisation, which illustrates the Duality Principle of Diebolt and Robert (1994). A first approximation embeds the finite support distribution on the latent variables within a continuous support distribution, easier to simulate by slice sampling, but we demonstrate later that the approximation can be very poor. We show that a genuine perfect slice sampler can be implemented for small sample sizes only and introduce an alternative perfect sampler based on a single backward chain, which can handle much larger sample sizes.
 
 
PERFECT SLICE SAMPLING FOR MIXTURES
CASELLA G., MENGERSEN K.L., ROBERT Christian P.
TITTERINGTON D.M.
2000-50
20-12-2000
 
Université de PARIS - DAUPHINE
Place du Maréchal de Lattre De Tassigny - 75775 PARIS CEDEX 16 - FRANCE
Téléphone : +33 (0)1 44-05-49-23 - fax : +33 (0)1 44-05-45-99