In jr-packages/efficientTutorial: Efficient R Programming

Parallel computing

Parallel computing

• Over the last few years, the speed of a single CPU core has remained static

local(source("code/08-hardware_cpu_speed.R", local=TRUE))

Parallel computing

• The parallel package comes with base R r library("parallel")
• To detect the number of cores on your machine, use r detectCores()

Quick Exercise: How many cores do you have?

Can my application be parallelised?

• An easy rule of thumb to determine if your for loop can be parallelised is to ask the question, can the loop be written in reverse and still give the same answer?
• For example {} for i in N:2 a[i] = a[i-1] + b[i] would result in a different a vector, or depending on how a has been defined, could crash your program.

Can my application be parallelised?

• However this loop {} for i in 1:N a[i] = monteCarloSimulation() results in identical results regardless of loop order

The apply family

The most common parallel applications to be parallelized are replacements of lapply, sapply, and apply

parLapply(cl, x, FUN, ...)
parApply(cl = NULL, X, MARGIN, FUN, ...)
parSapply(cl = NULL, X, FUN, ..., 
          simplify = TRUE, USE.NAMES = TRUE) 

mc = function(i, ...) runif(1)
N = 1

Example: Monte-Carlo simulation

• Suppose mc is a Monte-Carlo simulation function. Then we would have something like r results = numeric(N) for(i in 1:N) results[i] = mc(i)

Example: Monte-Carlo simulation

• Suppose mc is a Monte-Carlo simulation function. Then we would have something like r results = numeric(N) for(i in 1:N) results[i] = mc(i)
• We could re-write this for loop as r results = sapply(1:N, mc)
• Using sapply isn't faster than a standard for loop

Example: Monte-Carlo simulation

• To run in parallel is relatively straightforward.
• We make a cluster objects and specify the number of threads to generate r cl = makeCluster(2)
• Then change sapply to parSapply r results = parSapply(cl, 1:N, mc)
• and clean up r stopCluster(cl)

Exercise

vignette("parallel", package="efficientTutorial")

jr-packages/efficientTutorial documentation built on Feb. 16, 2020, 7:05 p.m.