Diagnostic checks to see if RCP groups are stable

For increasing size of hold-out samples, cooks distance and predictive log-likelihood are calculated and optionally plotted.

stability.regional_mix(
  model,
  oosSizeRange = NULL,
  times = model$n,
  mc.cores = 1,
  quiet = FALSE,
  doPlot = TRUE
)

Arguments

model	a regional_mix model, as obtained by the function regional_mix. This is the model whose stability is assessed. Model must contain titbits (see ?regional_mix and particular attention to the argument titbits=TRUE)
oosSizeRange	the size of the (successive) hold-out samples. If NULL (default), then a sequence of 10 sizes, from 1 to 0.2*model$n is used. The more numbers in this range, the slower the function will run.
times	the number of hold-out samples to use. If times=model$n and oosSize is 1, then the sample contains each and every site. Otherwise, it is a sample of size times from the possible combinations of possible hold-out sets.
mc.cores	the number of cores to farm the jobs out to.
quiet	should the progress bar be displayed (bar for each oosSizeRange)
doPlot	should the plots be produced? Default is that they should be.

Value

Produces a regional_mix_stab object. This is a list with the oosSizeRnage, disty (the mean Cook's Distance for each subset size), nRCP, n, predlogls (log-likelihood of out-of-sample sites), logl.sites (the in-sample log-likelihood for full data set).

Details

The plots produced are: 1) leave-some-out Cook's distance (see cooks.distance.regional_mix) against holdout sample size; and 2) the predictive log-likelihood for times sites, against the holdout sample size. In both plots, the values from the original model have been added to the plot.

Examples

if (FALSE) {
#not run as R CMD check complains about the time taken.
#This code will take a little while to run (about 3.5minutes on my computer)
 stability.regional_mix( fm, oosSizeRange=seq( from=1,to=fm$n,length=5),
                    times=fm$n, mc.cores=2, doPlot=FALSE);
}