dclone (version 1.2-0)

dc.parfit: Parallel model fitting with data cloning

Description

The iterative model fitting in on parallel workers with different number of clones.

Usage

dc.parfit(cl, data, params, model, inits, n.clones, 
multiply=NULL, unchanged=NULL, flavour = c("jags", "bugs"), ...)

Arguments

cl

A cluster object created by makeCluster.

data

A named list (not environment) containing the data.

params

Character vector of parameters to be sampled.

model

Character string (name of the model file), a function containing the model, or a or custommodel object (see Examples).

inits

Optional specification of initial values in the form of a list or a function (see Initialization at jags.model). If missing, will be treated as NULL and initial values will be generat

n.clones

An integer vector containing the numbers of clones to use itaratively.

multiply

Numeric or character index for list element(s) in the data argument to be multiplied by the number of clones instead of repetitions.

unchanged

Numeric or character index for list element(s) in the data argument to be left unchanged.

flavour

If "jags", the function jags.fit is called. If "bugs", the function bugs.fit is called.

...

Other values supplied to jags.fit, or bugs.fit, depending on the flavour argument. (Note, falvour = "bugs" has not be

Value

• An object inheriting from the class 'mcmc.list'.

Details

The dc.parfit is a parallel computing version of dc.fit. The vector n.clones is used to determine size balancing. If load balancing is also desired besides of size balancing (e.g. due to unequal performance of the workers, the option "dclone.LB" should be set to TRUE (by using options("dclone.LB" = TRUE)). By default, the "dclone.LB" option is FALSE for reproducibility reasons.

References

Lele, S.R., B. Dennis and F. Lutscher, 2007. Data cloning: easy maximum likelihood estimation for complex ecological models using Bayesian Markov chain Monte Carlo methods. Ecology Letters 10, 551--563.

See Also

Sequential version: dc.fit. Optimizing the number of workers: clusterSize, plotClusterSize.

Examples

set.seed(1234)
n <- 20
x <- runif(n, -1, 1)
X <- model.matrix(~x)
beta <- c(2, -1)
mu <- X %*% beta
Y <- rpois(n, exp(mu))
glm.model <- function() {
    for (i in 1:n) {
        Y[i] ~ dpois(lambda[i])
        log(lambda[i]) <- inprod(X[i,], beta[1,])
    }
    for (j in 1:np) {
        beta[1,j] ~ dnorm(0, 0.001)
    }
}
dat <- list(Y=Y, X=X, n=n, np=ncol(X))
k <- 1:3
cl <- makeCluster(2, type = "SOCK")
dcm <- dc.fit(dat, "beta", glm.model, n.clones=k, multiply="n", unchanged="np")
pdcm <- dc.parfit(cl, dat, "beta", glm.model, n.clones=k, multiply="n", unchanged="np")
summary(dcm)
summary(pdcm)
stopCluster(cl)

## Using WinBUGS/OpenBUGS
data(schools)
dat <- list(J = nrow(schools), y = schools$estimate, sigma.y = schools$sd)
bugs.model <- function(){
       for (j in 1:J){
         y[j] ~ dnorm (theta[j], tau.y[j])
         theta[j] ~ dnorm (mu.theta, tau.theta)
         tau.y[j] <- pow(sigma.y[j], -2)
       }
       mu.theta ~ dnorm (0.0, 1.0E-6)
       tau.theta <- pow(sigma.theta, -2)
       sigma.theta ~ dunif (0, 1000)
     }  
inits <- function(){
    list(theta=rnorm(nrow(schools), 0, 100), mu.theta=rnorm(1, 0, 100),
         sigma.theta=runif(1, 0, 100))
}
param <- c("mu.theta", "sigma.theta")
cl <- makeCluster(2, type = "SOCK")
sim2 <- dc.parfit(cl, dat, param, bugs.model, n.clones=1:2, 
    flavour="bugs", program="WinBUGS", multiply="J")
sim3 <- dc.parfit(cl, dat, param, bugs.model, n.clones=1:2, 
    flavour="bugs", program="OpenBUGS", multiply="J")
stopCluster(cl)