I am using the Atom editor to run MC simulations using Julia. I have implemented a parallel program for the same and am currently using addprocs(5). But, at the end or in the course of the simulations, the editor either hangs or crashes. Is there a way out of this ?
Run the REPL in a shell. In particular for long running calcs that is better anyway as otherwise your IDE-repl is blocked.
I am a relatively new user of Julia. If it is not too naive to ask, may I know what it means to run the REPL in a shell? I understand that the shell is for running system commands?
It means running julia outside from Atom in a shell. This may vary slightly depending on your operating system, but the basic idea is to start the julia binary. See, for instance, here.
Are you on windows? How did you install Julia?
I am using a Mac. I installed Julia from the downloaded installation file on their webpage.
So run Julia in a terminal app.
Oh, that was the shell reference in the earlier reply. Thank you !
I tried to run them in shell. However, the result is not any different. It hangs and takes unusually long to complete the simulation. If it is of any help, I run the following:
const mc = 1 ;
const mcd = 500 ; # Number of Monte Carlo iterations
theta = pmap(mc -> MonteCarlo(mc,mcf),1:1:mcf) ;
MonteCarlo is a function which on every run returns a tuple - 2 1500-by-100 arrays, 2 1500-by-100-by-100 matrices, 3 1-by-100 arrays. Currently I am trying with two workers,
Am I doing it right or is there a better way to implement it ? Thank you.
I’m sorry but I can’t make sense of the numbers you gave. These are huge, if true!
If you’re using
Float64 arrays, each iteration’s matrices will consume
2*1500*100*100*8=240 MB. With 500 runs, that comes out to 120 GB, so you may be exhausting your system’s memory. Is there any sparsity or simplification you could exploit?