runCVbatch.R

#!/usr/bin/env Rscript
# Aidan Draper
# June 12, 2018
# Batch system file for cross-validation of charcoal Bayesian model

# install.packages('spBayes', lib = "/afs/crc.nd.edu/user/a/adraper2/condor/Rlibs")
library(spBayes, lib = "/afs/crc.nd.edu/user/a/adraper2/condor/Rlibs")
library(mgcv)

args = commandArgs(trailingOnly=TRUE)

if (length(args)<2) {
  stop("Two arguments must be supplied (lake name, process number). Run again.", call.=FALSE)
} else if (length(args)==2) {
  # default output file
  lake.name = args[1]
  process = as.integer(args[2])
  if(process == 0){
    start = process + 1
    stop = process + 3
  } else {
    start = (process * 3) + 1
    stop = start + 2
  }
} else {
  stop("Too many arguments supplied. Run again.", call.=FALSE)
}

lake.inputs = paste("InputsCV", lake.name, ".rda",sep="")

load(lake.inputs)


###############################################################################
## Function created by Malcolm Itter
## May 25, 2018
###############################################################################
## RunAdaptMCMC_CV - Function to run MCMC simulation for univariate model
## using cross validation inputs
###############################################################################

run.cv = function(x){

  attach(x)

  out = try(adaptMetropGibbs(ltd=ltd,starting=starting,tuning=tune,batch=2400,report=400,
                             inputs=inputs,priors=priors))

  n.sim = 2400*25
  n.burn = 50000
  n.step = 5
  save.idx = seq(n.burn+1,n.sim,n.step)

  if (class(out)=="try-error"){
    samps = array(NA,dim=c(length(save.idx),length(starting)))
  } else {
    samps = out$p.theta.samples[save.idx,]
  }

  return(samps)

  detach(x)

}

####### END OF Malcolm's CODE ######


results = vector("list", ((stop - start) + 1))

for (i in start : stop){
  results[[i]] = run.cv(mod.inputs[[i]])
}

out.file = paste("OutCV",lake.name,process,".rda",sep="")

save(results,file=out.file)