CODEWITHSUNDEEP
rdata.table
persephone
persephone

Reputation: 420

rbinom (or else) by row and group in data.table

I want to draw Y for each row in group g in a data.table. To illustrate, take this simple simulation:

set.seed(123)

N = 50
g = 3

DT = data.table(id = rep(1:N,g),
                group_id = sort(rep(1:g, N)),
                p = runif(150, min = 0, max = 1)
                )

DT[, Y := rbinom(n = .N, size = 1, prob = p), by = group_id]  

DTs = split(DT, by = "group_id")

DTs = rbindlist(DTs[1], idcol=T)

DTs[, Y := rbinom(n = .N, size = 1, prob = p)]

head(DT)
head(DTs)

Why does Y differ for DT and DTs? I thought this to be equivalent.

> head(DT)
   id group_id         p Y
1:  1        1 0.2875775 1
2:  2        1 0.7883051 1
3:  3        1 0.4089769 0
4:  4        1 0.8830174 1
5:  5        1 0.9404673 1
6:  6        1 0.0455565 0


> head(DTs)
   .id id group_id         p Y
1:   1  1        1 0.2875775 1
2:   1  2        1 0.7883051 1
3:   1  3        1 0.4089769 1
4:   1  4        1 0.8830174 1
5:   1  5        1 0.9404673 1
6:   1  6        1 0.0455565 0

Upvotes: 0

Views: 230

Answers (1)

Allan Cameron
Allan Cameron

Reputation: 173928

The random number generator is shuffled every time you perform a random draw, so it is a different state at the point of the second binomial draw. If you want the same draw, you need to ensure the seed is set just before the draw:

set.seed(123)

N = 50
g = 3

DT = data.table(id = rep(1:N,g),
                group_id = sort(rep(1:g, N)),
                p = runif(150, min = 0, max = 1) # RNG changes state
                )

# Specify seed for binomial draw
set.seed(456)
DT[, Y := rbinom(n = .N, size = 1, prob = p), by = group_id]  # RNG changes state

DTs = split(DT, by = "group_id")

DTs = rbindlist(DTs[1], idcol=T)

# If we want the same draw, we need to set the seed to the same state
set.seed(456)
DTs[, Y := rbinom(n = .N, size = 1, prob = p)]

Giving you

head(DT)
#>    id group_id         p Y
#> 1:  1        1 0.2875775 0
#> 2:  2        1 0.7883051 1
#> 3:  3        1 0.4089769 1
#> 4:  4        1 0.8830174 1
#> 5:  5        1 0.9404673 1
#> 6:  6        1 0.0455565 0

head(DTs)
#>    .id id group_id         p Y
#> 1:   1  1        1 0.2875775 0
#> 2:   1  2        1 0.7883051 1
#> 3:   1  3        1 0.4089769 1
#> 4:   1  4        1 0.8830174 1
#> 5:   1  5        1 0.9404673 1
#> 6:   1  6        1 0.0455565 0

Upvotes: 1

Related Questions