examples with future and batchtools for PBS Torque

batchtools.torque.tmpl

#!/bin/bash
#PBS -N <%= job.name %>
#PBS -o <%= log.file %>
#PBS -l walltime=<%= resources$walltime %>,mem=<%= resources$memory %>M
#PBS -q batch
#PBS -j oe

## setup modules

## create our own temp dir (and clean it up later)
#mkdir /data/${USER}/${USER}-${PBS_JOBID}
#export TMPDIR=/data/${USER}/${USER}-${PBS_JOBID}

<%= sprintf("export OMP_NUM_THREADS=%i", resources$omp.threads) -%>
<%= sprintf("export OPENBLAS_NUM_THREADS=%i", resources$blas.threads) -%>
<%= sprintf("export MKL_NUM_THREADS=%i", resources$blas.threads) -%>

## export value of DEBUGME environemnt var to slave
#export DEBUGME=<%= Sys.getenv("DEBUGME") %>

## run R
Rscript -e 'batchtools::doJobCollection("<%= uri %>")'

## Cleanup
#rm -rf /data/${USER}/${USER}-${PBS_JOBID}

hpc_example_r_batchtools_future.R

library(magic)
library(future)
library(batchtools)
library(future.batchtools)

# Example 1: submit jobs to cluster via future
plan(batchtools_torque, resources = list(walltime = '00:05:00', memory = '2Gb')) # specify that the PBS Torque cluster is used and the resources we want

get_magig_diagsum <- function(i){
  magic_matrix <- magic(as.integer(i))
  sum(diag(magic_matrix))
}

single_result %<-% get_magig_diagsum(5) # submit a single job
while (!resolved(single_result)) Sys.sleep(5) # wait until the job is finished

single_result

# Example 2: submit many jobs in parallel via batchtools
reg = makeRegistry(file.dir = '.future_registry', seed = 1) # creates a folder for a registry with all neccessary bookkeeping info

reg$packages <- c( 'magic') # indicates required packages
reg$cluster.functions = makeClusterFunctionsTORQUE(template = "batchtools.torque.tmpl") # indicates that 

res <- btlapply(5:9, fun = get_magig_diagsum, reg = reg,resources = list(walltime = '00:05:00', memory = '2Gb')) # syntax is similar to the standard R lapply function

res

clearRegistry() # clean up
removeRegistry()

# Example 3: parallelize brms execution

library(brms)
p <- read.csv("https://stats.idre.ucla.edu/stat/data/poisson_sim.csv")
p <- within(p, {
  prog <- factor(prog, levels=1:3, labels=c("General", "Academic", 
                                            "Vocational"))
  id <- factor(id)
})

plan(list(
  tweak(batchtools_torque, resources = list(walltime = '00:20:00', memory = '6Gb', packages = c('brms'))),
  tweak(batchtools_torque, resources = list(walltime = '00:05:00', memory = '6Gb', packages = c('brms')))
), .cleanup = T) # two-level paralellization setup to compile on the cluster and then run the chains in parallel

brm_job <- future({
  brm(num_awards~math+prog, data = p, family = 'poisson', future = T)},
  packages = c('brms'), globals = c('p')) # the job is submitted to compile on the cluster and the individual chains are in turn started in parallel

while (!resolved(brm_job)) Sys.sleep(5) # wait until the job is finished

brm_val <- value(brm_job) # load the results
brm_val