dclone (version 2.3-2)

update.mcmc.list: Automatic updating of an MCMC object from JAGS

Description

Automatic updating of an MCMC object until a desired statistic value reached.

Usage

updated.model(object, ...)
# S3 method for mcmc.list
update(object, fun,
    times = 1, n.update = 0, n.iter, thin, ...)

Value

updated.model returns the state of the JAGS model after updating and sampling. This can be further updated by the function update.jags

and sampled by coda.samples if convergence diagnostics were not satisfactory.

update returns an MCMC object with "updated.model" attribute.

Arguments

object

A fitted MCMC object ('mcmc.list' class for example), with "updated.model" attribute.

fun

A function that evaluates convergence of the MCMC chains, must return logical result. See Examples. The iterative updating quits when return value is TRUE. Can be missing, in which case there is no stopping rule.

times

Number of times the updating should be repeated. If fun returns TRUE, updating is finished and MCMC object is returned.

n.update

Number of updating iterations. The default 0 indicates, that only n.iter iterations are used.

n.iter

Number of iterations for sampling and evaluating fun. If missing, value is taken from object.

thin

Thinning value. If missing, value is taken from object.

...

Other arguments passed to coda.samples.

Author

Peter Solymos,

Details

updated.model can be used to retrieve the updated model from an MCMC object fitted via the function jags.fit and dc.fit (with flavour = "jags"). The update method is a wrapper for this purpose, specifically designed for the case when MCMC convergence is problematic. A function is evaluated on the updated model in each iteration of the updating process, and an MCMC object is returned when iteration ends, or when the evaluated function returns TRUE value.

n.update and n.iter can be vectors, if lengths are shorter then times, values are recycled.

Data cloning information is preserved.

See Also

jags.fit, coda.samples, update.jags

Examples

if (FALSE) {
## simple regression example from the JAGS manual
jfun <- function() {
    for (i in 1:N) {
        Y[i] ~ dnorm(mu[i], tau)
        mu[i] <- alpha + beta * (x[i] - x.bar)
    }
    x.bar <- mean(x[])
    alpha ~ dnorm(0.0, 1.0E-4)
    beta ~ dnorm(0.0, 1.0E-4)
    sigma <- 1.0/sqrt(tau)
    tau ~ dgamma(1.0E-3, 1.0E-3)
}
## data generation
set.seed(1234)
N <- 100
alpha <- 1
beta <- -1
sigma <- 0.5
x <- runif(N)
linpred <- crossprod(t(model.matrix(~x)), c(alpha, beta))
Y <- rnorm(N, mean = linpred, sd = sigma)
## list of data for the model
jdata <- list(N = N, Y = Y, x = x)
## what to monitor
jpara <- c("alpha", "beta", "sigma")
## fit the model with JAGS
regmod <- jags.fit(jdata, jpara, jfun, n.chains = 3)
## get the updated model
upmod <- updated.model(regmod)
upmod
## automatic updating
## using R_hat < 1.1 as criteria
critfun <- function(x)
    all(gelman.diag(x)$psrf[,1] < 1.1)
mod <- update(regmod, critfun, 5)
## update just once
mod2 <- update(regmod)
summary(mod)
}