Random number generation

Dirk Eddelbuettel — Dec 24, 2012 | source

R comes with several random number generators supporting the ability to draw random samples from a wide variety of statistical distributions.

Rcpp builds on this, and provides access to the same random number generators, and distributions. Moreover, thanks to Rcpp sugar, these can be accessed in a vectorised manner, as we illustrated in the post simulating pi.

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
NumericMatrix rngCpp(const int N) {
  RNGScope scope;		// ensure RNG gets set/reset
  NumericMatrix X(N, 4);
  X(_, 0) = runif(N);
  X(_, 1) = rnorm(N);
  X(_, 2) = rt(N, 5);
  X(_, 3) = rbeta(N, 1, 1);
  return X;
}

 set.seed(42)     # setting seed
 M1 <- rngCpp(5)
 M1

       [,1]     [,2]    [,3]    [,4]
[1,] 0.9148  0.04788  2.9967 0.79234
[2,] 0.9371 -1.10460  0.8067 0.38822
[3,] 0.2861  0.53902 -1.7226 0.56423
[4,] 0.8304  0.58021  0.8337 0.02646
[5,] 0.6417 -0.65750  0.4019 0.04242

 set.seed(42)	  # resetting seed
 M2 <- cbind( runif(5), rnorm(5), rt(5, 5), rbeta(5, 1, 1))
 M2

       [,1]     [,2]    [,3]    [,4]
[1,] 0.9148  0.04788  2.9967 0.79234
[2,] 0.9371 -1.10460  0.8067 0.38822
[3,] 0.2861  0.53902 -1.7226 0.56423
[4,] 0.8304  0.58021  0.8337 0.02646
[5,] 0.6417 -0.65750  0.4019 0.04242

 
 all.equal(M1, M2)

[1] TRUE

The other method of using the R random-number generator is in scalar mode, one variable and draw at a time. This is very similar to the description of this API in the Writing R Extensions manual, and provided by Rcpp in the R namespace:

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
NumericVector rngCppScalar() {
  RNGScope scope;		// ensure RNG gets set/reset
  NumericVector x(4);
  x[0] = R::runif(0,1);
  x[1] = R::rnorm(0,1);
  x[2] = R::rt(5);
  x[3] = R::rbeta(1,1);
  return(x);
}

 set.seed(42)
 v1 <- rngCppScalar()
 v1

[1] 0.9148 1.5307 1.0510 0.8653

 set.seed(42)
 v2 <-c(runif(1), rnorm(1,0,1), rt(1,5), rbeta(1,1,1))
 v2

[1] 0.9148 1.5307 1.0510 0.8653

 all.equal(v1, v2)

[1] TRUE

tags: rmath rng sugar

Using RcppProgress to control the long computations in C++ — Karl Forner
An accept-reject sampler using RcppArmadillo::sample() — Jonathan Olmsted
Using the RcppArmadillo-based Implementation of R's sample() — Christian Gunning and Jonathan Olmsted

Random number generation

Related Articles