stan.fit {dclone}

Fit Stan models with cloned data

Description

Convenient functions designed to work well with cloned data arguments and Stan.

Usage

stan.fit(data, params, model, inits = NULL,
    seed = sample.int(.Machine$integer.max, 1),
    n.chains = 3,
    format = c("mcmc.list", "stanfit"),
    stan.model = TRUE, fit = NA, ...)
stan.model(object, ...)
stan.parfit(cl, data, params, model, inits = NULL,
    seed = sample.int(.Machine$integer.max, n.chains),
    n.chains = 3,
    format = c("mcmc.list", "stanfit"),
    stan.model = TRUE, fit = NA, ...)

Arguments

data

A list (or 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 custommodel object.

inits

Optional specification of initial values in the form of a list or a function. If NULL, initial values will be generated automatically.

seed

Random seed.

n.chains

number of Markov chains.

format

Desired output format.

stan.model

Logical, if stanmodel object should be returned.

fit

Fitted Stan object.

cl

A cluster object created by makeCluster, or an integer, see parDosa and evalParallelArgument.

object

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

...

Further arguments.

Value

By default, an stan.fit returns an mcmc.list object. If data cloning is used via the data argument, summary returns a modified summary containing scaled data cloning standard errors (scaled by sqrt(n.clones)), and R_{hat} values (as returned by gelman.diag).

stan.model returns the stanmodel object.

stan.parfit runs chains using multiple cores when cl is an integer. Using a cluster object leads to recompiling the model (therefore fit is ignored), and might not be very quick to run.

Author(s)

Peter Solymos, [e-mail]

See Also

Underlying functions: stan and stanfit in package rstan

Methods: dcsd, confint.mcmc.list.dc, coef.mcmc.list, quantile.mcmc.list, vcov.mcmc.list.dc

Examples

## Not run: 
if (require(rstan)) {
    model <- custommodel("data {
          int<lower=0> N;
          vector[N] y;
          vector[N] x;
        }
        parameters {
          real alpha;
          real beta;
          real<lower=0> sigma;
        }
        model {
          alpha ~ normal(0,10);
          beta ~ normal(0,10);
          sigma ~ cauchy(0,5);
          for (n in 1:N)
            y[n] ~ normal(alpha + beta * x[n], sigma);
        }")
    N <- 100
    alpha <- 1
    beta <- -1
    sigma <- 0.5
    x <- runif(N)
    y <- rnorm(N, alpha + beta * x, sigma)
    dat <- list(N=N, y=y, x=x)
    params <- c("alpha", "beta", "sigma")
    ## compile on 1st time only
    fit0 <- stan.fit(dat, params, model)
    ## reuse compiled fit0
    fit <- stan.fit(dat, params, model, fit=fit0)
    sm <- stan.model(fit)
    summary(fit)
    sm

    ## data cloning
    dcdat <- dclone(dat, n.clones=2, multiply="N")
    dcfit <- stan.fit(dcdat, params, model, fit=fit0)
    summary(dcfit)
    nclones(dcfit)

    ## using parallel options
    cl <- makeCluster(2)
    ## cannot utilize compiled fit0
    fit2 <- stan.parfit(cl=cl, dat, params, model)
    stopCluster(cl)
    if (.Platform$OS.type != "windows") {
        ## utilize compiled fit0
        fit3 <- stan.parfit(cl=2, dat, params, model, fit=fit0)
    }
}

## End(Not run)