| Title: | Ensemble Sampler for Affine-Invariant MCMC |
| Version: | 3.2.0 |
| Description: | Provides ensemble samplers for affine-invariant Monte Carlo Markov Chain, which allow a faster convergence for badly scaled estimation problems. Two samplers are proposed: the 'differential.evolution' sampler from ter Braak and Vrugt (2008) <doi:10.1007/s11222-008-9104-9> and the 'stretch' sampler from Goodman and Weare (2010) <doi:10.2140/camcos.2010.5.65>. |
| License: | GPL-2 |
| URL: | https://hugogruson.fr/mcmcensemble/,https://github.com/Bisaloo/mcmcensemble |
| BugReports: | https://github.com/Bisaloo/mcmcensemble/issues |
| Depends: | R (≥ 3.5) |
| Imports: | future.apply, progressr |
| Suggests: | bayesplot, coda, mockery, testthat (≥ 3.0.0), knitr,rmarkdown, mvtnorm |
| Encoding: | UTF-8 |
| RoxygenNote: | 7.3.2 |
| Config/testthat/edition: | 3 |
| Config/Needs/website: | spacefillr |
| VignetteBuilder: | knitr |
| NeedsCompilation: | no |
| Packaged: | 2025-07-24 08:02:11 UTC; hugo |
| Author: | Hugo Gruson [cre, aut, cph], Sanda Dejanic [aut, cph], Andreas Scheidegger [aut, cph] |
| Maintainer: | Hugo Gruson <hugo.gruson+R@normalesup.org> |
| Repository: | CRAN |
| Date/Publication: | 2025-07-24 08:40:17 UTC |
MCMCEnsembleSampler
Description
This package implements a particle Monte Carlo Markov Chain samplerwith two different ways of creating proposals.
Author(s)
Maintainer: Hugo Grusonhugo.gruson+R@normalesup.org (ORCID) [copyright holder]
Authors:
Sanda Dejanic [copyright holder]
Andreas Scheidegger (ORCID) [copyright holder]
See Also
Useful links:
Report bugs athttps://github.com/Bisaloo/mcmcensemble/issues
MCMC ensemble sampler
Description
Ensemble Markov Chain Monte Carlo sampler with different strategies togenerate proposals. Either thestretch move as proposed by Goodman andWeare (2010), or adifferential evolution jump move similar to Braak andVrugt (2008).
Usage
MCMCEnsemble( f, inits, max.iter, n.walkers = 10 * ncol(inits), method = c("stretch", "differential.evolution"), coda = FALSE, ...)Arguments
f | function that returns a single scalar value proportional to the logprobability density to sample from. |
inits | A matrix (or data.frame) containing the starting values for thewalkers. Each column is a variable to estimate and each row is a walker |
max.iter | maximum number of function evaluations |
n.walkers | number of walkers (ensemble size). An integer greater than |
method | method for proposal generation, either |
coda | logical. Should the samples be returned ascoda::mcmc.listobject? (defaults to |
... | further arguments passed to |
Value
if
coda = FALSEa list with:samples: A three dimensional array of samples with dimensions
walkerxgenerationxparameterlog.p: A matrix with the log density evaluate for each walker at eachgeneration.
if
coda = TRUEa list with:samples: A object of classcoda::mcmc.list containing all samples.
log.p: A matrix with the log density evaluate for each walker at eachgeneration.
In both cases, there is an additional attribute (accessible viaattr(res, "ensemble.sampler")) recording which ensemble sampling algorithmwas used.
References
ter Braak, C. J. F. and Vrugt, J. A. (2008) Differential Evolution MarkovChain with snooker updater and fewer chains. Statistics and Computing, 18(4),435–446,doi:10.1007/s11222-008-9104-9
Goodman, J. and Weare, J. (2010) Ensemble samplers with affine invariance.Communications in Applied Mathematics and Computational Science, 5(1), 65–80,doi:10.2140/camcos.2010.5.65
Examples
## a log-pdf to sample fromp.log <- function(x) { B <- 0.03 # controls 'bananacity' -x[1]^2/200 - 1/2*(x[2]+B*x[1]^2-100*B)^2}## set options and starting pointn_walkers <- 10unif_inits <- data.frame( "a" = runif(n_walkers, 0, 1), "b" = runif(n_walkers, 0, 1))## use stretch moveres1 <- MCMCEnsemble(p.log, inits = unif_inits, max.iter = 150, n.walkers = n_walkers, method = "stretch")attr(res1, "ensemble.sampler")str(res1)## use stretch move, return samples as 'coda' objectres2 <- MCMCEnsemble(p.log, inits = unif_inits, max.iter = 150, n.walkers = n_walkers, method = "stretch", coda = TRUE)attr(res2, "ensemble.sampler")summary(res2$samples)plot(res2$samples)## use different evolution move, return samples as 'coda' objectres3 <- MCMCEnsemble(p.log, inits = unif_inits, max.iter = 150, n.walkers = n_walkers, method = "differential.evolution", coda = TRUE)attr(res3, "ensemble.sampler")summary(res3$samples)plot(res3$samples)