apply.parallel.R

#############################
#### Apply functions     ####
#############################
matrixData = matrix(1:40, nrow = 10, ncol = 4)

### Median of rows
## Using for loop
rowm = rep(NA, nrow(matrixData))
for(r in 1:nrow(matrixData)) {
  rowm[r] = median(matrixData[r,])
}
rowm

## Using apply
apply(matrixData, 1, median)

## Apply on columns instead
apply(matrixData, 2, median)

## Apply on lists
listData = list(a = rnorm(10), b = rnorm(10), c = rnorm(10))
lapply(listData, mean)
lapply(listData, function(x) x * 100)

## Try not to return a list, but a vector/matrix
sapply(listData, mean)
sapply(listData, function(x) x * 100)

## Apply on multiple lists
listData2 = list(d = 1:10, e = 11:20, f = 21:30)
formatText = function(x, y) {
  paste(round(x, 2), y, sep=", ")
}
mapply(formatText, listData, listData2)

#############################
#### Parallell execution ####
#############################
bigdata = lapply(1:10^4, function(x) rnorm(10))

### Multicore ###
library(multicore)
options(mc.cores = 10) ## Set number of cores through options

## Do something non-parallel
system.time({
  bigresult = lapply(bigdata, median)
})

system.time({
  bigresult = mclapply(bigdata, median)
})

### Foreach ###
library(foreach)

## No parallel execution
system.time({
  bigresult = foreach(d = bigdata) %do% median(d)
})

## Parallel execution
## Multiple cores
library(doMC) ## Load parallel backend
registerDoMC(cores = 10)

system.time({
  bigresult = foreach(d = bigdata) %dopar% median(d)
})

## Other backends available in do* packages (e.g. doSNOW for SNOW clusters)

## Different .combine function
bigmatrix = foreach(d = bigdata, .combine = cbind) %dopar% d / 2