How can foreach() in the parallel R package be handled a repeat loop with breaks?

问题

I have written a routine that takes considerable time without parallelization.

The issue is that I am unsure what to iterate over, since I have a repeat loop with breaks.

The loop consists of the following code snippet (for loop not shown):

repeat{
    if(R < p){
        HAC.sim(K = K, N = ceiling(Nstar), Hstar = Hstar, probs = probs, perms = perms, equal.freq = FALSE, subset.haps = NULL)
} else{
    break
  }
}

I would like to use foreach() with the parallel backend; however I am not certain what is needed for

foreach(i = 1:???){
    some code
}

since, I do not know ahead of time when the repeat loop will stop.

回答1:

You can iterate on only the number of cores you have. Then you can detect with memory-mapping when one core has found the solution and then stop the others.

library(bigstatsr)
library(foreach)

ncores <- nb_cores()  # or parallel::detectCores() - 1
fbm <- FBM(1, 1, init = 0)   # shared memory
p <- 0.9999
HAC.sim <- function() runif(1)

cl <- parallel::makeCluster(ncores)
doParallel::registerDoParallel(cl)

res <- foreach(i = 1:ncores, .combine = 'c') %dopar% {
  repeat {
    if (fbm[1] != 0) return(NULL)
    R <- HAC.sim()
    if (R >= p) {
      fbm[1] <- 1   # tell the others to stop
      return(R)
    } 
  }
}

parallel::stopCluster(cl)

res

来源：https://stackoverflow.com/questions/47019114/how-can-foreach-in-the-parallel-r-package-be-handled-a-repeat-loop-with-breaks