fast sampling in R

Question

Is there a faster method for taking a random sub sample (without replacement), than the base::sample function?

Answer 1

you can get a little bit of a speed-up by eliminating the base::sample function call:

> x<- rnorm(10000)
> system.time(for(i in 1:100000) x[.Internal(sample(10000L, 10L, FALSE, NULL))])
   user  system elapsed 
  2.873   0.017   2.851 
> system.time(for(i in 1:100000) sample(x,10))
   user  system elapsed 
  3.420   0.025   3.258 

Depending on your problem there may be other more clever ways of speeding up your code. Think about ways to replace many small calls to sample with one big one.

Answer 2

no

I can get 10,000 samples in 3 ms on my laptop with replacement. Without replacement I can get them in 5ms. Drawing multiple times from 500 distributions it takes 66 ms. How fast did you need it to be?

Answer 3

The dqrng package tackles faster sampling in R. Here is one example & benchmark:

library(dqrng)
library(bench)

m <- 1000
n <- 99999
all <- m * n
bm <- bench::mark(samp = sample(x = c(1, -1), size = all, replace = TRUE),
                  dqsamp = dqsample(x = c(1,-1), size = all, replace = TRUE),
                  check = FALSE, 
                  iterations = 3)
bm

# # A tibble: 2 x 13
#   expression      min   median `itr/sec` mem_alloc `gc/sec` n_itr  n_gc total_time
#   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl> <int> <dbl>   <bch:tm>
# 1 samp          6.37s    6.59s     0.153    1.12GB    0.153     3     3     19.56s
# 2 dqsamp        1.07s    1.43s     0.723    1.12GB    0.482     3     2      4.15s
# # ... with 4 more variables: result <list>, memory <list>, time <list>, gc <list>

Here is a related blogpost: https://www.r-bloggers.com/2019/04/fast-sampling-support-in-dqrng/.