Clever Geek Handbook

R: Parallel optimization DEoptim - number of cores

I am trying to use an optimization package DEoptim

in R with a constant optimization problem, and since my cost function is taking a long time to evaluate (2 min), I am trying to use parallel computation. My questions:

  • What's the difference between parameters paralleltype=1

    ( parallel

    ) and 2

    ( foreach

    )? When to use one of the two?

  • Is it possible to specify the number of cores with paralleltype=1

    so as not to take all available cores for calculation (for example, 50 cores on 64 cores)?

+3


source to share


2 answers


ads. 1 - This is explained in the package documentation.

ads. 2 - It takes some fiddling with two functions: DEoptim.control and DEoptim

First, add a variable - say limitCores - to the DEoptim.control function that controls the parameters for executing DE optimization.



Second, modify the DEoptim wrapper function to act on the limit set by limitCores.

if (ctrl$parallelType == 1) {

    if (!is.na(ctrl$limitCores)) {

        if (ctrl$limitCores<1) useCores <- round(parallel::detectCores()*ctrl$limitCores) else useCores <- ctrl$limitCores

        cl <- parallel::makeCluster(parallel::detectCores())

    } else {

        cl <- parallel::makeCluster(parallel::detectCores())

    }

Full code: http://pastebin.com/NumDx4ae

+1


source


Thanks for your code, mjaniec!

I know it's been a long time since you posted it, but I think on line 91 it should be:

cl <- parallel::makeCluster(limitCores)


instead:

cl <- parallel::makeCluster(parallel::detectCores())
0


source






More articles:


All Articles