parallel :: clusterExport如何从全局环境传递嵌套函数? [英] parallel::clusterExport how to pass nested functions from global environment?

问题描述

我正在制作一个函数(myFUN)，该函数一次调用parallel :: parApply，并提供一个函数yourFUN作为参数.

在许多情况下，yourFUN将包含来自全局环境的自定义函数.

因此，尽管我可以将"yourFUN"传递给parallel :: clusterExport，但我无法事先知道其中的函数名称，而clusterExport因为找不到它们而向我返回错误.

我不想导出yourFUN的整个封闭环境，因为它可能很大.

我是否可以仅导出运行yourFUN所需的变量?

实际功能很长，下面是该错误的最小示例:

mydata <- matrix(data = 1:9, 3, 3)

perfFUN <- function(x) 2*x

opt_perfFUN <- function(y) max(perfFUN(y))

avg_perfFUN <- function(w) perfFUN(mean(w))

myFUN <- function(data, yourFUN, n_cores = 1){

  cl <- parallel::makeCluster(n_cores)
  parallel::clusterExport(cl, varlist = c("yourFUN"), envir = environment())

  parallel::parApply(cl, data, 1, yourFUN)

}

myFUN(data = mydata, yourFUN = opt_perfFUN)
myFUN(data = mydata, yourFUN = avg_perfFUN)

 Error in checkForRemoteErrors(val) : one node produced an error: could not find function "perfFUN" 

非常感谢您！

解决方案

一种可能的解决方案，请使用:

myFUN <- function(data, yourFUN, n_cores = 1) {

  cl <- parallel::makeCluster(n_cores)
  on.exit(parallel::stopCluster(cl), add = TRUE)

  envir <- environment(yourFUN)
  parallel::clusterExport(cl, varlist = ls(envir), envir = envir)

  parallel::parApply(cl, data, 1, yourFUN)  
}

I'm making a function (myFUN) that calls parallel::parApply at one point, with a function yourFUN that is supplied as an argument.

In many situations, yourFUN will contain custom functions from the global environment.

So, while I can pass "yourFUN" to parallel::clusterExport, I cannot know the names of functions inside it beforehand, and clusterExport returns me an error because it cannot find them.

I don't want to export the whole enclosing environment of yourFUN, since it might be very big.

Is there a way for me to export only the variables necessary for running yourFUN?

The actual function is very long, here is a minimized example of the error:

mydata <- matrix(data = 1:9, 3, 3)

perfFUN <- function(x) 2*x

opt_perfFUN <- function(y) max(perfFUN(y))

avg_perfFUN <- function(w) perfFUN(mean(w))

myFUN <- function(data, yourFUN, n_cores = 1){

  cl <- parallel::makeCluster(n_cores)
  parallel::clusterExport(cl, varlist = c("yourFUN"), envir = environment())

  parallel::parApply(cl, data, 1, yourFUN)

}

myFUN(data = mydata, yourFUN = opt_perfFUN)
myFUN(data = mydata, yourFUN = avg_perfFUN)

 Error in checkForRemoteErrors(val) : one node produced an error: could not find function "perfFUN" 

Thank you very much!

解决方案

A possible solution, use:

myFUN <- function(data, yourFUN, n_cores = 1) {

  cl <- parallel::makeCluster(n_cores)
  on.exit(parallel::stopCluster(cl), add = TRUE)

  envir <- environment(yourFUN)
  parallel::clusterExport(cl, varlist = ls(envir), envir = envir)

  parallel::parApply(cl, data, 1, yourFUN)  
}

这篇关于parallel :: clusterExport如何从全局环境传递嵌套函数?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！