Section 12 Optimisation of R code

12.1 Vectorised code

• Identify the part of code that can take advantage of logical test, subsetting and element-wise execution. That part of code certainly can be vectorised.

• Many pre-existing R fucntions are already vectorised and optimised to perform efficiently. Many of these functions are actually written in C and Fortran lanugages.

• If a combination of if and for loop is required, it is more likely that the code could be vectorised.

• Explore if you can use the apply functions which are optimised to run faster.

• Identify the operations that could be done outside the for loop.

• Define the storage size of the vector before executing the for loop.

12.2 Example 1

abs_loop <- function(x){
  for(i in 1:length(x)){
    if(x[i] < 0){
        x[i] <- x[i]*(-1)
    }
  }
  return(x)
}

abs_vec <- function(x){
  x[x < 0] * (-1)
  return(x)
}


a <- rep(c(-1,1), length=5000000)

system.time(abs_loop(a))

system.time(abs_vec(a))

system.time(abs(a))

• Note: R built-in function abs is optimised to execute the operation of calculating absolute values more efficiently.

12.3 Example 2

add_loop1 <- function(x){
  z <- rep(NA, length=length(x))
  for(i in 1:length(x)){
      z[i] <- x[i] + 1
  }
  return(z)
}

add_loop2 <- function(x){
  z <- NA
  for(i in 1:length(x)){
      z[i] <- x[i] + 1
  }
  return(z)
}


a <- rep(c(1:10), length=5000000)

system.time(add_loop1(a))

system.time(add_loop2(a))