Is there a systematic way to convert R code with loops to vectorized code?

Question

Most looping code looks like this

retVal=NULL
for i {
  for j {
    result <- *some function of vector[i] and vector[j]* 
    retVal = rbind(retVal,result)
  }
}

Since this is so common, is there a systematic way of translating this idiom?

Can this be extended to most loops?

Answer 1

The first goal should to get working code. You are there. Then try some simple optmizations. E.g.

retVal <- matrix(NA, ni, nj)     # assuming your result is scalar
for (i in 1:ni)
   for (j in 1:nj)
       retVal[i][j] <- *some function of yours*

will already run much faster as you do not reallocate memory for each i,j combination.

As for the looping, you can start by replacing the inner loop with something from the apply family. I am not aware of something fully general to answer your question -- it depends what arguments your function takes and what type of return object it produces.

Answer 2

The plyr package provides a set of general tools for replacing looping constructs when you're work with a big data structure by breaking it into pieces, processing each piece independently and then joining the results back together.