Resampling Markov Chain Monte Carlo Algorithms: Basic Analysis and Empirical Comparisons

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Sampling from complex distributions is an important but challenging topic in scientific and statistical computation. We synthesize three ideas, tempering, resampling, and Markov moving, and propose a general framework of resampling Markov chain Monte Carlo (MCMC). This framework not only accommodates various existing algorithms, including resample-move, importance resampling MCMC, and equi-energy sampling, but also leads to a generalized resample-move algorithm. We provide some basic analysis of these algorithms within the general framework, and present three simulation studies to compare these algorithms together with parallel tempering in the difficult situation where new modes emerge in the tails of previous tempering distributions. Our analysis and empirical results suggest that generalized resample-move tends to perform the best among all the algorithms studied when the Markov kernels lead to fast mixing or even locally so toward restricted distributions, whereas parallel tempering tends to perform the best when the Markov kernels lead to slow mixing, without even converging fast to restricted distributions. Moreover, importance resampling MCMC and equi-energy sampling perform similarly to each other, often worse than independence Metropolis resampling MCMC. Therefore, different algorithms seem to have advantages in different settings.

Original languageEnglish (US)
Pages (from-to)328-356
Number of pages29
JournalJournal of Computational and Graphical Statistics
Volume24
Issue number2
DOIs
StatePublished - Apr 3 2015

Fingerprint

Markov Chain Monte Carlo Algorithms
Resampling
Markov Chain Monte Carlo
Parallel Tempering
Tend
kernel
Analysis of Algorithms
Energy
Tail
Markov chain Monte Carlo
Simulation Study
Framework
Sampling

All Science Journal Classification (ASJC) codes

  • Statistics and Probability
  • Discrete Mathematics and Combinatorics
  • Statistics, Probability and Uncertainty

Keywords

  • Equi-energy sampling
  • Importance resampling
  • Parallel tempering
  • Potts model
  • Resample-move
  • Sequential Monte Carlo

Cite this

@article{0d4d1257f45945d9ae347bd789b3824a,
title = "Resampling Markov Chain Monte Carlo Algorithms: Basic Analysis and Empirical Comparisons",
abstract = "Sampling from complex distributions is an important but challenging topic in scientific and statistical computation. We synthesize three ideas, tempering, resampling, and Markov moving, and propose a general framework of resampling Markov chain Monte Carlo (MCMC). This framework not only accommodates various existing algorithms, including resample-move, importance resampling MCMC, and equi-energy sampling, but also leads to a generalized resample-move algorithm. We provide some basic analysis of these algorithms within the general framework, and present three simulation studies to compare these algorithms together with parallel tempering in the difficult situation where new modes emerge in the tails of previous tempering distributions. Our analysis and empirical results suggest that generalized resample-move tends to perform the best among all the algorithms studied when the Markov kernels lead to fast mixing or even locally so toward restricted distributions, whereas parallel tempering tends to perform the best when the Markov kernels lead to slow mixing, without even converging fast to restricted distributions. Moreover, importance resampling MCMC and equi-energy sampling perform similarly to each other, often worse than independence Metropolis resampling MCMC. Therefore, different algorithms seem to have advantages in different settings.",
keywords = "Equi-energy sampling, Importance resampling, Parallel tempering, Potts model, Resample-move, Sequential Monte Carlo",
author = "Zhiqiang Tan",
year = "2015",
month = "4",
day = "3",
doi = "10.1080/10618600.2014.897625",
language = "English (US)",
volume = "24",
pages = "328--356",
journal = "Journal of Computational and Graphical Statistics",
issn = "1061-8600",
publisher = "American Statistical Association",
number = "2",

}

Resampling Markov Chain Monte Carlo Algorithms : Basic Analysis and Empirical Comparisons. / Tan, Zhiqiang.

In: Journal of Computational and Graphical Statistics, Vol. 24, No. 2, 03.04.2015, p. 328-356.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Resampling Markov Chain Monte Carlo Algorithms

T2 - Basic Analysis and Empirical Comparisons

AU - Tan, Zhiqiang

PY - 2015/4/3

Y1 - 2015/4/3

N2 - Sampling from complex distributions is an important but challenging topic in scientific and statistical computation. We synthesize three ideas, tempering, resampling, and Markov moving, and propose a general framework of resampling Markov chain Monte Carlo (MCMC). This framework not only accommodates various existing algorithms, including resample-move, importance resampling MCMC, and equi-energy sampling, but also leads to a generalized resample-move algorithm. We provide some basic analysis of these algorithms within the general framework, and present three simulation studies to compare these algorithms together with parallel tempering in the difficult situation where new modes emerge in the tails of previous tempering distributions. Our analysis and empirical results suggest that generalized resample-move tends to perform the best among all the algorithms studied when the Markov kernels lead to fast mixing or even locally so toward restricted distributions, whereas parallel tempering tends to perform the best when the Markov kernels lead to slow mixing, without even converging fast to restricted distributions. Moreover, importance resampling MCMC and equi-energy sampling perform similarly to each other, often worse than independence Metropolis resampling MCMC. Therefore, different algorithms seem to have advantages in different settings.

AB - Sampling from complex distributions is an important but challenging topic in scientific and statistical computation. We synthesize three ideas, tempering, resampling, and Markov moving, and propose a general framework of resampling Markov chain Monte Carlo (MCMC). This framework not only accommodates various existing algorithms, including resample-move, importance resampling MCMC, and equi-energy sampling, but also leads to a generalized resample-move algorithm. We provide some basic analysis of these algorithms within the general framework, and present three simulation studies to compare these algorithms together with parallel tempering in the difficult situation where new modes emerge in the tails of previous tempering distributions. Our analysis and empirical results suggest that generalized resample-move tends to perform the best among all the algorithms studied when the Markov kernels lead to fast mixing or even locally so toward restricted distributions, whereas parallel tempering tends to perform the best when the Markov kernels lead to slow mixing, without even converging fast to restricted distributions. Moreover, importance resampling MCMC and equi-energy sampling perform similarly to each other, often worse than independence Metropolis resampling MCMC. Therefore, different algorithms seem to have advantages in different settings.

KW - Equi-energy sampling

KW - Importance resampling

KW - Parallel tempering

KW - Potts model

KW - Resample-move

KW - Sequential Monte Carlo

UR - http://www.scopus.com/inward/record.url?scp=84931043581&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84931043581&partnerID=8YFLogxK

U2 - 10.1080/10618600.2014.897625

DO - 10.1080/10618600.2014.897625

M3 - Article

AN - SCOPUS:84931043581

VL - 24

SP - 328

EP - 356

JO - Journal of Computational and Graphical Statistics

JF - Journal of Computational and Graphical Statistics

SN - 1061-8600

IS - 2

ER -