Journal article
Langevin-Type Models II: Self-Targeting Candidates for MCMC Algorithms
Methodology and computing in applied probability, Vol.1(3), p.307
10/01/1999
DOI: 10.1023/A:1010090512027
Abstract
Byline: O. Stramer (1), R. L. Tweedie (2) Keywords: Hastings algorithms; Metropolis algorithms; Markov chain Monte Carlo; diffusions; Langevin models; discrete approximations; posterior distributions; irreducible Markov processes; geometric ergodicity; uniform ergodicity; Gibbs sampling The Metropolis-Hastings algorithm for estimating a distribution [pi] is based on choosing a candidate Markov chain and then accepting or rejecting moves of the candidate to produce a chain known to have [pi] as the invariant measure. The traditional methods use candidates essentially unconnected to [pi]. We show that the class of candidate distributions, developed in Part I (Stramer and Tweedie 1999), which "self-target" towards the high density areas of [pi], produce Metropolis-Hastings algorithms with convergence rates that appear to be considerably better than those known for the traditional candidate choices, such as random walk. We illustrate this behavior for examples with exponential and polynomial tails, and for a logistic regression model using a Gibbs sampling algorithm. The detailed results are given in one dimension but we indicate how they may extend successfully to higher dimensions. Author Affiliation: Article History: Registration Date: 30/09/2004
Details
- Title: Subtitle
- Langevin-Type Models II: Self-Targeting Candidates for MCMC Algorithms
- Creators
- O Stramer - University of IowaR. L Tweedie - University of Minnesota
- Resource Type
- Journal article
- Publication Details
- Methodology and computing in applied probability, Vol.1(3), p.307
- Publisher
- Springer
- DOI
- 10.1023/A:1010090512027
- ISSN
- 1387-5841
- eISSN
- 1573-7713
- Language
- English
- Date published
- 10/01/1999
- Description audience
- Academic
- Academic Unit
- Statistics and Actuarial Science
- Record Identifier
- 9984257722302771
Metrics
10 Record Views