[R] Fixed size permutations

[Peter Ehlers]

Nick Fankhauser wrote:
> I'm using functions to return a matrix of all permutations of a
> specified size from a larger list for predictor selection.
> For each predictor size I use a seperate function like this:
> 
> bag2 <- function(n) {
>     rl <- c()
>     for (i1 in seq(n)) {
>         for (i2 in seq(n)) {
>             if (length(unique(c(i1,i2)))==1) {next}
>             rl <- cbind(rl,matrix(c(i1,i2)))
>         }
>     }
>     rl
> }
> 
> bag3 <- function(n) {
>     rl <- c()
>     for (i1 in seq(n)) {
>         for (i2 in seq(n)) {
>             for (i3 in seq(n)) {
>                 if (length(unique(c(i1,i2,i3)))==1) {next}
>                 rl <- cbind(rl,matrix(c(i1,i2,i3)))
>             }
>         }
>     }
>     rl
> }
> 
> But I think it should be somehow possible in R to use one general
> function for all sizes. Can someone help?
> I don't exactly know how this kind of permutation is called, maybe this
> would help to find a solution.
> 
> Nick
> 
Looks to me like you're just doing expand.grid, but without
the 'all variables equal' cases. If that's correct and if
n > 1, then this might do it:

bag <- function(n,K){
   L <- vector(mode='list', length=K)
   L <- lapply(L, function(x) {x <- seq_len(n)})
   d <- expand.grid(L)
   keeprow <- apply(d, 1, function(x) {var(x)!=0})
     ## I'm blanking out on a better way to test for equal values
   td <- t(as.matrix(d[keeprow,]))
   dimnames(td) <- c(NULL, NULL)
   rtd <- td[nrow(td):1,]
   rtd
}
all.equal(bag(5,2), bag2(5))
#[1] TRUE

all.equal(bag(9,3), bag3(9))
#[1] TRUE

  -Peter Ehlers