internal .paraRasterFun not running in parallel on Windows

[bleutner]

note to self:

• everything with doParallel backend (.parXapply) is fine
• only .paraRasterFun apparently doesn't cooperate (Windows only), e.g. predict in superClass

[16EAGLE]

.paraRasterFun itself called from global environment worked fine on two different Windows machines, using the available CPUs as expected. However, there appears to be a problem when using superClass related to the caret package. On Windows, running

beginCluster()
sc <- superClass(x, trainData = train, responseCall = "class",
                 model = "rf", tuneLength = 1, trainPartition = 0.7)

causes

Error in e$fun(obj, substitute(ex), parent.frame(), e$data) : 
  unable to find variable "optimismBoot"

This does not appear if beginCluster() is not called. The issue is known by caret users when trying to run caret functions such as train in parallel (issue: topepo/caret#706). Users report that this error also occured on Linux systems. Running the above commands using the current fixed github version of caret with devtools::install_github('topepo/caret/pkg/caret') works fine, however, this version seems not to be on CRAN yet due to some downward dependencies and the issue is still open.

Seems that just waiting for caret to be updated would solve the issue related to superClass. Until now, I could not reproduce further parallelization issues on windows.

[Baitcookie]

On windows (better stop using this) I was successful using caret::train in parallel with parallel::makeForkCluster.