-1

I have a data set that I need to split into three mutually exclusive random samples of different sizes. I used: 

testdata<-sample(47959,14388,replace=FALSE,prob=NULL)

to try and create a sample (the data set size is 47959) but I don't know how to make this sample into a manipulatable data set in R.

m0nhawk
  • 22,980
  • 9
  • 45
  • 73
pdhis95
  • 3
  • 2

1 Answers1

2

Some data:

set.seed(42)
x <- sample(20, size=100, replace=TRUE)
head(x)
## [1] 19 19  6 17 13 11

Random Sizes

Create an index of all 1:3, and use that to subset your data:

i <- sample(1:3, size=length(x), replace=TRUE)
head(i)
## [1] 2 1 1 2 3 3

Now break it into the three groups (many ways to do this):

x.grouped <- split(x, i)
str(x.grouped)
## List of 3
##  $ 1: int [1:31] 19 6 15 20 9 5 8 9 18 20 ...
##  $ 2: int [1:30] 19 17 14 10 6 10 19 3 10 19 ...
##  $ 3: int [1:39] 13 11 15 3 15 19 12 2 8 19 ...

The relative sizes of the three groups will vary randomly.

Controlled sizes

indices represents the size you desire in each group.

indices <- c(20, 50, 30)
indices.cs <- cumsum(indices)
x.unsorted <- sample(x)
xs.grouped.sized <- mapply(function(a,b) x.unsorted[a:b],
    1+lag(indices.cs, default=0),
    indices.cs,
    SIMPLIFY=FALSE)
str(xs.grouped.sized)
## List of 3
##  $ : int [1:20] 2 7 13 1 19 7 14 20 19 1 ...
##  $ : int [1:50] 13 6 19 4 19 20 20 11 17 3 ...
##  $ : int [1:30] 1 10 7 16 9 16 17 11 14 8 ...

Edit: Updated Implementation

indices <- sample(rep(1:3, times = c(20,50,30)))
str(split(x, indices))
## List of 3
##  $ 1: int [1:20] 6 3 10 6 10 20 17 8 5 13 ...
##  $ 2: int [1:50] 19 19 17 15 14 15 19 20 3 19 ...
##  $ 3: int [1:30] 13 11 15 19 10 12 3 11 14 1 ...
r2evans
  • 141,215
  • 6
  • 77
  • 149
  • Maybe instead of `lapply(1:3, function(j) x[i == j])`, try just `split(x, i)`. – Thomas Mar 28 '15 at 20:32
  • True, victim of habit and quick hacks (which is why I tend to caveat some answers with "many ways to do this"). In my head, I've always grouped `split` with `by` which produces structures that take a little different care. I see now that was an error. Thanks @Thomas! – r2evans Mar 28 '15 at 20:36
  • you could do the controlled sizes like this instead: `str(split(sample(1:sum(indices <- c(20, 50, 30))), rep(seq_along(indices), times = indices)))` – rawr Mar 28 '15 at 21:13
  • ... and then use that list of vectors to index the original data, sure. One liners are always fun (to write, not to debug/maintain!). My updated implementation above should certainly *not* use magic constants (`1:3`), ala `sample(rep(seq_along(indices <- c(20,50,30)), times=indices))`. – r2evans Mar 28 '15 at 21:19