emcmc: Eruptive MCMC

In thibautjombart/fancysamplers: fancy ways to sample from complicated densities

Description

This is a variation on the tempering algorithm. It relies on alternative blocks of cold chains with blocks of very hot chains.

Usage

emcmc(f, param = list(x = 0), to.move = TRUE, n = 200,
  cold.block.size = 200, hot.block.size = 100, sd = 1, min.temp = 10,
  max.temp = 100)

Arguments

f	a function computing log-densities
param	a list of named parameters passed on to 'f'; parameters must be given their initial values
to.move	a logical or an integer vector indicating which parts of 'param' should be moved in the MCMC
n	a number of iterations for the eMCMC
cold.block.size	the size of cold blocks
hot.block.size	the size of hot blocks
sd	the standard deviation of the proposal normal distribution
min.temp	the maximum temperature to be used
max.temp	the maximum temperature to be used

Author(s)

Thibaut Jombart [e-mail]

Examples

## try with a basic normal density
f1 <- function(x){dnorm(x, log=TRUE)}

## basic MCMC
mc1 <- metro(f1)
plot(mc1)

## eMCMC
emc1 <- emcmc(f1)
plot(emc1[[1]]) # all chains
plot(emc1[[2]]) # thinned

## try with a mixture of densities
fmix <- function(x){log(dnorm(x, mean=-5) +
                        dnorm(x, mean=5, sd=.5))}

plot(function(x) exp(fmix(x)), xlim=c(-10,10))
title("Target density")

## basic MCMC
mc2 <- metro(fmix, list(x=0), n=6e4)
plot(mc2)

## eMCMC
emc2 <- emcmc(fmix)
plot(emc2[[1]]) # all chains
plot(emc2[[2]]) # thinned

thibautjombart/fancysamplers documentation built on May 31, 2019, 9:56 a.m.