Move diag calculations within convergence test

[mikegilchrist]

Code appears to evaluate costly diag <- coda::geweke.diag(mcmcobj, frac1 = frac1, frac2 = frac2) even when results are not used.

• Issue likely exists in other convergence.test functions.
• Also note unnecessary defining of trace

Current code

convergence.test.Rcpp_MCMCAlgorithm
function (object, samples = 10, frac1 = 0.1, frac2 = 0.5, thin = 1, 
    plot = FALSE, what = "Mutation", mixture = 1) 
{
    trace <- object$getLogPosteriorTrace()
    loglik.trace <- object$getLogPosteriorTrace()
    trace.length <- length(loglik.trace)
    start <- max(1, trace.length - samples)
    mcmcobj <- coda::mcmc(data = loglik.trace[start:trace.length], 
        thin = thin)
    diag <- coda::geweke.diag(mcmcobj, frac1 = frac1, frac2 = frac2)
    if (plot) {
        coda::geweke.plot(mcmcobj, frac1 = frac1, frac2 = frac2)
    }
    else {
        return(diag)
    }
}

Proposed code

convergence.test.Rcpp_MCMCAlgorithm
function (object, samples = 10, frac1 = 0.1, frac2 = 0.5, thin = 1, 
    plot = FALSE, what = "Mutation", mixture = 1) 
{
    loglik.trace <- object$getLogPosteriorTrace()
    trace.length <- length(loglik.trace)
    start <- max(1, trace.length - samples)
    mcmcobj <- coda::mcmc(data = loglik.trace[start:trace.length], 
        thin = thin)
    if (plot) {
        coda::geweke.plot(mcmcobj, frac1 = frac1, frac2 = frac2)
    }
    else {
        diag <- coda::geweke.diag(mcmcobj, frac1 = frac1, frac2 = frac2)
        return(diag)
    }
}

[mikegilchrist]

Looking at R/traceObject.R, it appears that it doesn't need to be fixed.

convergence.test.Rcpp_Trace <- function(object, samples = 10, frac1 = 0.1,
                                        frac2 = 0.5, thin = 1, plot = FALSE, what = "Mutation", mixture = 1)
{
.
.
.
mcmcobj <- coda::mcmc(data=current.trace, start=start, thin=thin)
  if(plot){
    coda::geweke.plot(mcmcobj, frac1=frac1, frac2=frac2)
  } else{
    diag <- coda::geweke.diag(mcmcobj, frac1=frac1, frac2=frac2)
    return(diag)
  }
}