Avoid two for loops in R

Question

I have a R code that can do convolution of two functions...

convolveSlow <- function(x, y) {  
nx <- length(x); ny <- length(y)  
xy <- numeric(nx          


        

Answer 1

As Dirk says, compiled code can be a lot faster. I had to do this for one of my projects and was surprised at the speedup: ~40x faster than Andrie's solution.

> a <- runif(10000)
> b <- runif(10000)
> system.time(convolveFast(a, b))
   user  system elapsed 
  7.814   0.001   7.818 
> system.time(convolveC(a, b))
   user  system elapsed 
  0.188   0.000   0.188 

I made several attempts to speed this up in R before I decided that using C code couldn't be that bad (note: it really wasn't). All of mine were slower than Andrie's, and were variants on adding up the cross-product appropriately. A rudimentary version can be done in just three lines.

convolveNotAsSlow <- function(x, y) {
  xyt <- x %*% t(y)
  ds <- row(xyt)+col(xyt)-1
  tapply(xyt, ds, sum)
}

This version only helps a little.

> a <- runif(1000)
> b <- runif(1000)
> system.time(convolveSlow(a, b))
   user  system elapsed 
  6.167   0.000   6.170 
> system.time(convolveNotAsSlow(a, b))
   user  system elapsed 
  5.800   0.018   5.820 

My best version was this:

convolveFaster <- function(x,y) {
  foo <- if (length(x)=0 & bar.rc

This was quite a bit better, but still not nearly as fast as Andrie's

> system.time(convolveFaster(a, b))
   user  system elapsed 
  0.280   0.038   0.319