biasing conditional sample

-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
project.org] On Behalf Of Rui Barradas
Sent: Sunday, November 11, 2012 4:36 PM
To: dms at riseup.net
Cc: r-help at r-project.org
Subject: Re: [R] biasing conditional sample

Hello,

The function that follows returns a matrix, not a data.frame but does
what you ask for.


fun <- function(x, y, n){
     f <- function(x, y){
         while(TRUE){
             rnd <- sample(x, 1)
             if(!any(rnd %in% y)) break
         }
         rnd
     }
     for(i in seq_len(n)){
         tmp <- apply(y, 1, function(.y) f(x, .y))
         y <- cbind(y, tmp)
     }
     y
}


a <- cbind(sample(1:10, 100, TRUE)) # must have dims
fun(1:10, a, 4)  # returns 5 columns, 'a' plus 4


Hope this helps,

Rui Barradas
Em 11-11-2012 19:06, dms at riseup.net escreveu:

Hi all,

I'm looking for some help to bias the sample function. Basically, I'd

like

to generate a data frame where the first column is completely random,

the

second, however, is conditional do the first, the third is

conditional to

the first and the second and so on. By conditional I mean that I

shouldn't

have repeated values in the line. I know it could be easily

implemented

using permutation, but it is not the case here. I need at least five
columns. Any idea to achieve what do I need?


set.seed(51)
  data <- data.frame(
      id=as.factor(1:100),
      a=as.factor(sample(1:10, size=100, replace=TRUE)),
      b=as.factor(sample(1:10, size=100, replace=TRUE)),
      c=as.factor(sample(1:10, size=100, replace=TRUE)),
      d=as.factor(sample(1:10, size=100, replace=TRUE)),
      e=as.factor(sample(1:10, size=100, replace=TRUE))
)

guide.html