0

I'm attempting to parallelize a loop for conducting manual 10-fold cross validation with RF. However, what happens is only that the memory of the pc overloads and I'm forced to quit R. I don't understand what the problem might be, the code seems correct to me:

library(parallel);library(doParallel)
library(foreach)

#Setting trainControl function
control <- trainControl(method="none")

#Creating folders randomly for 10-fold CV
set.seed(123)
folds <- createFolds(data$y, k=10, list=FALSE)
data <- cbind(data, folds)

#Creating matrix for MAE
MAE = matrix(NA, nrow=10, ncol=1)
colnames(MAE)=c("MAE")

#Parallized CV loop
cl<-makeCluster(7)
registerDoParallel(cl)

  foreach(n=1:10) %dopar% {
  library(caret)
  library(randomForest)
  train <- data[data$folds!=n,]; train$folds=NULL
  test <- data[data$folds==n,]; test$folds=NULL
  set.seed(123)
  rf= train(y~., data=train, 
             method="rf", 
             trControl=control)
  pred <- predict(object = rf, newdata = test, type= "raw")
  library(Metrics)
  mae <- mae(test$y,pred)
  MAE[n,] = c(mae)
  }
  
stopCluster(cl)

Thank you for any suggestion!

kris
  • 109
  • 6
  • 2
    First of all, your `foreach` loop is not setup correctly. You cannot assign into the matrix in the global environment from within parallelized code. Study the [vignette](https://cran.r-project.org/web/packages/doParallel/vignettes/gettingstartedParallel.pdf). Then, if you parallelize a task usually you need more memory. That can quickly become an issue and you need to balance speed gains with memory demand. That means you need to optimize the number of cores (or get more RAM). – Roland Mar 08 '23 at 12:14

0 Answers0