在 Rcpp 中获取随机种子

Question

I need to minimize a function, let's say f(theta) . In f(theta) , I draw VERY BIG matrices from a given distribution. So, throughout the optimization process, I have to keep the same matrices. However, as the matrices are too big, I cannot save them. Thus, for every value of theta , I decide to find a way to draw the same matrices. This can be done if I define a seed, which will be used each time the function is called. But I do not want the user to set the seed, I need the software to get the current seed and then use that seed for every new value of theta .

This a simplify example if I use R.

f <- function(theta, SEED) {
  .Random.seed <- SEED
  # I simplify the problem, I replace the big matrices by a small vector
  return(sum((rnorm(10) - theta)^2))
}

optim(par = 0, fn = f, method = "Brent", lower = -1, upper = 1, SEED = .Random.seed)

.Random.seed allows me to get a seed which will be used in f . This ensures that, during the optimization, rnorm(10) will not be changed although it is not saved. How can I construct a similar function using Rcpp ?

Answer 1

Rcpp already deal with this for you, and that is (amply) documented. To wit:

> set.seed(42)
> rnorm(3)
[1]  1.370958 -0.564698  0.363128
> rnorm(3)
[1]  0.632863  0.404268 -0.106125
> Rcpp::cppFunction("Rcpp::NumericVector myrn(int n) {return Rcpp::rnorm(3);}")
> myrn(3)   # __not reseeded, different__
[1]  1.511522 -0.094659  2.018424
> set.seed(42)
> myrn(3)   # reseeded, same as above from R
[1]  1.370958 -0.564698  0.363128
> myrn(3)   # ditto
[1]  0.632863  0.404268 -0.106125
> 

After you call the function you are supposed to call, set.seed() , you get identical draws in both R and C++. So call a seed before calling your function and all is well.