Lancaster EPrints

Filtering Methods for Mixture Models .

Fearnhead, P and Meligkotsidou, L (2007) Filtering Methods for Mixture Models . Journal of Computational and Graphical Statistics, 16 (3). pp. 586-607.

[img]
Preview
PDF (Mixture_RDS.pdf)
Download (2711Kb) | Preview

    Abstract

    We consider Bayesian inference for mixture distributions of known number of components via a set of filtering recursions. We extend a method - proposed in an earlier article - of direct simulation for discrete mixture distributions in order to analyze continuous mixture models. Furthermore, we introduce resampling steps similar to those in particle filters within the steps of the filtering recursions, which make calculations efficient and enable us to analyze larger datasets. The proposed algorithm for "resampled direct simulation" is a generalization of the particle filter which allows for merging identical/similar particles prior to resampling. We compare the proposed algorithm with this particle filter and with the Gibbs sampler using simulated data and real datasets.

    Item Type: Article
    Journal or Publication Title: Journal of Computational and Graphical Statistics
    Uncontrolled Keywords: Direct Simulation ; Gibbs Sampling ; Importance Sampling ; Particle Filters ; Perfect Simulation ; Rejection Sampling
    Subjects: UNSPECIFIED
    Departments: Faculty of Science and Technology > Mathematics and Statistics
    ID Code: 746
    Deposited By: Prof Paul Fearnhead
    Deposited On: 08 Nov 2007
    Refereed?: Yes
    Published?: Published
    Last Modified: 09 Oct 2013 15:41
    Identification Number:
    URI: http://eprints.lancs.ac.uk/id/eprint/746

    Actions (login required)

    View Item