0

How can I randomly sample e.g. three groups within a data.table so that the result contains three groups with all rows from the original data.table? 

library(data.table)
dat <- data.table(ids=1:20, groups=sample(x=c("A","B","C", "D", "E", "F"), 20, 
replace=TRUE))

I know how to select 10 rows randomly from a data.table:

dat.sampl1 <- as.data.table(sapply(dat[], sample, 10))

And also how to sample by group 

dat[,.SD[sample(.N, min(.N,3))], by = groups]

But how to randomly sample groups? So the result should look like:

    ids groups
     1      F
    11      F
     3      F
    18      F
     8      A
     9      A
    10      A
    17      A
    19      A
    12      E
    14      E
    16      E
Henrik
  • 65,555
  • 14
  • 143
  • 159
user2147915
  • 1
  • 1
  • 2
    I don't understand what you are asking. If you are going to use `sample()`, then use `set.seed()` so your data is reproducible. It seems you have some constraint so it's not a simple random sample. Is this some sort of conditional sampling perhaps? – MrFlick May 14 '18 at 18:03
  • Does this answer your question? [Sample random rows within each group in a data.table](https://stackoverflow.com/questions/16289182/sample-random-rows-within-each-group-in-a-data-table) – Union find Jul 07 '23 at 19:42

2 Answers2

5

Do you mean something like:

set.seed(123)
dat <- data.table(ids=1:20, groups=sample(x=c("A","B","C", "D", "E", "F"), 20, 
replace=TRUE))
dat[groups %in% sample(unique(dat[, groups]), size = 3)][order(groups)]
#     ids groups
#  1:   3      C
#  2:  10      C
#  3:  12      C
#  4:   7      D
#  5:   9      D
#  6:  14      D
#  7:   4      F
#  8:   5      F
#  9:   8      F
# 10:  11      F
# 11:  16      F
# 12:  20      F

If you want to sample groups with replacement, you can do the following, where A has been sampled twice:

dat[unique(dat[, list(groups)])[sample(.N, 3, replace = TRUE)], on = "groups"]
#    ids groups
# 1:   3      C
# 2:  10      C
# 3:  12      C
# 4:   6      A
# 5:  15      A
# 6:  18      A
# 7:   6      A
# 8:  15      A
# 9:  18      A
Weihuang Wong
  • 12,868
  • 2
  • 27
  • 48
1

This code works, using a single line of base R code using %in% to check an index which is generated using the sample function:

df1[df1[,'groups'] %in% sample(unique(df1[,'groups']), size = 3, replace = F), ]

For example:

> df1 <- data.frame("ids" = 1:20, "groups" = sample(LETTERS[1:4], size = 20, replace = T))
> df2 <- df1[df1[,'groups'] %in% sample(unique(df1[,'groups']), size = 3, replace = F), ]
> df2[order(df2[,'groups']),]
   ids groups
4    4      B
6    6      B
18  18      B
20  20      B
1    1      C
2    2      C
3    3      C
9    9      C
12  12      C
16  16      C
19  19      C
7    7      D
11  11      D
rg255
  • 4,119
  • 3
  • 22
  • 40