Examples of different parallelisations in R

parallel_examples.R

# https://nceas.github.io/oss-lessons/parallel-computing-in-r/parallel-computing-in-r.html

library(parallel)
library(foreach)
library(doParallel)
library(snow)
set.seed(42)

n_subs <- 100000
n_trials <- 5000

mu <- rnorm(n_subs, 800, 50)
sigma <- rnorm(n_subs, 180, 10)
tau <- runif(n_subs, 1, 100)

data <- data.frame(
  mu = mu, 
  sigma = sigma, 
  tau = tau)




# function that will be called in parallel
do_sim <- function(parameters, n_trials = 5000){
  
  sub_parms <- as.numeric(parameters)
  
  sub_rts <- rnorm(n_trials, sub_parms[1], sub_parms[2]) + 
    rexp(rexp(n_trials, 1 / sub_parms[3]))
  
  sub_cdfs <- round(quantile(sub_rts, probs = c(.1, .3, .5, .7, .9)), 0)
  
  out <- rbind(data.frame(), sub_cdfs)
  
  sub_cdfs
  
}


do_reg <- function(){
  # pass the data to a set of lists
  split_data <- split(data, 1:nrow(data))
  out_reg <- lapply(split_data, do_sim)
  out_reg <- matrix(unlist(out_reg), ncol = 5, byrow = TRUE)
}

# parallel package
do_multi <- function(){
  num_cores <- detectCores()
  split_data <- split(data, 1:nrow(data))
  out_multi <- mclapply(split_data, do_sim, mc.cores = num_cores)
  out_multi <- matrix(unlist(out_multi), ncol = 5, byrow = TRUE)
}

# foreach package
do_for_each <- function(){
  num_cores <- detectCores()
  registerDoParallel(num_cores)
  out_for_each <- foreach(i = 1:nrow(data)) %dopar% {
    do_sim(data[i, ])
  }
  out_for_each<- matrix(unlist(out_for_each), ncol = 5, byrow = TRUE)
}

# snow package (https://hernanresnizky.com/2014/01/10/quick-guide-to-parallel-r-with-snow/)
do_snow <- function(){
  num_cores <- detectCores()
  clus <- makeCluster(num_cores)
  clusterExport(clus, "n_trials")
  split_data <- split(data, 1:nrow(data))
  out_snow <- parRapply(clus, data, do_sim)
  out_snow <- matrix(out_snow, ncol = 5, byrow = FALSE)
}

system.time(out_reg <- do_reg())
system.time(out_multi <- do_multi())
system.time(out_for_each <- do_for_each())
system.time(out_snow <- do_snow())