How to randomise a vector and keep the frequency of the elements fixed?

Question

Extending this former question, how can I shuffle (randomize) the following vector

a1 = c(1, 1, 2, 2, 2, 2, 3, 3, 4, 5, 5, 5)

in order to get something like this:

a2 = c(5, 5, 3, 3, 3, 3, 1, 1, 2, 4, 4, 4)

or even better like this:

a3 = c(4, 4, 4, 2, 3, 3, 3, 3, 1, 1, 5, 5)?

such that each element could randomly change to another but with keeping the number of each element constant?

Seems a promising idea! I'll also give it a try to see if I can start from there and produce a3. Thanks! — Rotail, Oct 22 '16 at 03:51
@42-, Matlab does not use commas to separate array elements. I know because I am a Matlab refugee. — R. Schifini, Oct 22 '16 at 04:04
Alas, some language uses square-brackets and that was the basis of my guess. — IRTFM, Oct 22 '16 at 04:15
@42- I'm an R citizen exiled to MATLAB by a boss for a while :P — Rotail, Oct 22 '16 at 14:40

score 4 · Accepted Answer · answered Oct 22 '16 at 03:54

4

You can try something like this: create a factor from a1 with randomly shuffled levels and then convert it to integers:

as.integer(factor(a1, levels = sample(unique(a1), length(unique(a1)))))
# [1] 5 5 4 4 4 4 3 3 2 1 1 1

answered Oct 22 '16 at 03:54

Psidom

209,562
33
339
356

This produces a2 not a3, yet. – Rotail Oct 22 '16 at 03:58
Do the same elements have to be adjacent to each other after the second shuffle? – Psidom Oct 22 '16 at 04:00
Or maybe `t <- table(a1); rep(names(t), sample(t))`,`[1] "1" "2" "2" "3" "3" "3" "3" "4" "4" "4" "5" "5"` if you don't care about the run length encoding. – Psidom Oct 22 '16 at 04:12

IRTFM · Answer 2 · 2016-10-22T04:18:09.223

3

Seems like a perfect application for rle and its inverse rep:

rand_inverse_rle <- function(x) { x=sort(x)
   ord=sample (length(rle(x)$values) )  
   unlist( mapply( rep, rle(x)$values[ord], rle(x)$lengths[ord]))}
rand_inverse_rle(a1)
#----------
 [1] 3 3 4 5 5 5 2 2 2 2 1 1

This was my reading of a function needed to satisfy the natural language requirements:

>  a1 = sample( c(1, 1, 2, 2, 2, 2, 3, 3, 4, 5, 5, 5) )
> a1
 [1] 5 2 5 2 5 1 3 4 2 2 3 1
> rand_inverse_rle(a1)
 [1] 5 5 5 4 2 2 2 2 3 3 1 1
> rand_inverse_rle(a1)
 [1] 1 1 3 3 5 5 5 2 2 2 2 4
> rand_inverse_rle(a1)
 [1] 1 1 3 3 4 5 5 5 2 2 2 2

edited Oct 22 '16 at 04:18

answered Oct 22 '16 at 03:58

IRTFM

258,963
21
364
487

1

i don't know whether sorting is needed. It _was_ needed for my reading of the English language description, even if the example had already been sorted.. addendum added. – IRTFM Oct 22 '16 at 04:14

score 3 · Answer 3 · answered Oct 22 '16 at 04:00

The data:

a1 <- c(1, 1, 2, 2, 2, 2, 3, 3, 4, 5, 5, 5)

First steps:

# extract values and their frequencies
val <- unique(a1)

tab <- table(a1)
freq <- tab[as.character(val)]

Keep original order of frequencies but sample values

rep(sample(val), freq)
# [1] 4 4 1 1 1 1 3 3 5 2 2 2

Keep original frequencies but sample order of values

rep(sa <- sample(val), freq[as.character(sa)])
# [1] 4 2 2 2 2 3 3 1 1 5 5 5

How to randomise a vector and keep the frequency of the elements fixed?

3 Answers3

Keep original order of frequencies but sample values

Keep original frequencies but sample order of values