Using counter from loop when using foreach() - r

Question

I have a large simulation that I would like to run on multiple cores.
For this I am using the foreach() package.
I am iterating a loop 1000 times and within the loop I am using the counter from the loop as a position vector, for exampe:
reps<-1000
for (i in reps){
a[i]<-mean(rnorm(100))
}
If I do the same thing with foreach:
library(foreach)
cl<-makeCluster(8)
registerDoParallel(cl)
ls<-foreach(icount(reps)) %dopar% {
rnorm(100)
}
I can no longer use the current counter i as in the original loop.
Is there a way to use it?
I am also fine with having a counter, like everytime one iteration passes I do i=i+1 starting from i=0.

Answer 1

As Roland suggests, you could write the "foreach" version as:
reps <- 1000
ls <- foreach(i=icount(reps), .combine='c') %dopar% {
mean(rnorm(100))
}
The variable i isn't used, but it's available if you want it in the future.
Using "for" loops inside "foreach" loops can be very useful for getting better performance, since it can decrease the number of iterations and allow the workers to do more of the work in parallel:
reps <- 1000
ls <- foreach(n=rep(reps/8, 8), .combine='c') %dopar% {
a <- numeric(n)
for (i in seq_len(n)) {
a[i] <- mean(rnorm(100))
}
a
}
It's fine to make assignments to a inside the "foreach" loop because it's a local variable. When the "for" loop finishes, a is returned, and finally all of the a vectors are combined with the "c" function by the master.
Notice that this example is using the original "for" loop inside the "foreach" loop. This is more efficient because the master doesn't have to do nearly as much work to send out the tasks and collect and combine the results.
Actually, I would use "sapply" instead of the "for" loop and "idiv" (from the "iterators" package) instead of "rep":
reps <- 1000
ls <- foreach(n=idiv(reps, chunks=8), .combine='c') %dopar% {
sapply(seq_len(n), function(i) mean(rnorm(100)))
}
With "idiv" I don't have to worry about what will happen if the number of iterations is not evenly divisible by the number of workers.