CODEWITHSUNDEEP
rforeachparallel-processingdata.table
statquant
statquant

Reputation: 14400

loading files in parallel not working with foreach + data.table

I would like to use foreach in conjuction with data.table (v.1.8.7) to load files and bind them. foreach is not parallelizing, and returning a warning...

write.table(matrix(rnorm(5e6),nrow=5e5),"myFile.csv",quote=F,sep=",",row.names=F,col.names=T) 
library(data.table); 
#I use fread from data.table 1.8.7 (dev) for performance and useability
DT = fread("myFile.csv")

Now suppose I have n of those files to load and rowbind, I would like to parralellize it. (I am on Windows, so no forking)

allFiles = rep("myFile.csv",4) # you can change 3 to whatever

using lapply

f1 <- function(allFiles){
    DT <- lapply(allFiles, FUN=fread) #will load sequentially myFile.csv 3 times with fread
    DT <- rbindlist(DT);
    return(DT);
}

single thread

using parallel (part of R as 2.14.0)

library(parallel)
f2 <- function(allFiles){
    mc <- detectCores(); #how many cores?
    cl <- makeCluster(mc); #build the cluster
    DT <- parLapply(cl,allFiles,fun=fread); #call fread on each core (well... using each core at least)
    stopCluster(cl);
    DT <- rbindlist(DT);
    return(DT);
}

here it uses both proc

now I want to use foreach

library(foreach)
f3 <- function(allFiles){
    DT <- foreach(myFile=allFiles, .combine='rbind', .inorder=FALSE) %dopar% fread(myFile)
    return(DT);
}

enter image description here

Here are some benchmarks confirming I can't kave foreach working 

system.time(DT <- f1(allFiles));
utilisateur     systÞme      ÚcoulÚ
      34.61        0.14       34.84
system.time(DT <- f2(allFiles));
utilisateur     systÞme      ÚcoulÚ
       1.03        0.40       24.30    
system.time(DT <- f3(allFiles));
executing %dopar% sequentially: no parallel backend registered
utilisateur     systÞme      ÚcoulÚ
      35.05        0.22       35.38

Upvotes: 4

Views: 2636

Answers (1)

Roland
Roland

Reputation: 132989

Just to get this answered:

As the warning message tells you, there is no parallel backend registered for foreach. Read this vignette to learn how to do that.

Simple example from the vignette:

library(doParallel) 
cl <- makeCluster(3) 
registerDoParallel(cl) 
foreach(i=1:3) %dopar% sqrt(i)

Upvotes: 2

Related Questions