setRNG (version 2013.9-1)

setRNG-package: setRNG

Description

Programs to set random number generator (and seed) in R and S.

Usage

library("setRNG")

Introduction

This library provides tools to simplify recording and resetting the random number generator, to help make monte carlo experiments easily reproducible. It uses the R/S tools for setting the seed, but also records and sets the mechanism for converting uniform numbers to normally distributed numbers. (It could be extended to other transformations, but I have not done that.)

The setRNG function would typically be called by simulation programs (see example) to set the RNG information if given, and record the RNG information in all cases. This information can be returned with the result of the simulation. That way the simulation can always be reproduced if necessary.

The library also implements an approach to random number generation which allows the same random experiments to be replicated in S and R. The functions in the S/ directory allow the R results using Wichmann-Hill and Box-Muller to be replicated in S. These were done with the aid of an example from B. D. Ripley. (The files in the S/ directory of the package are for use with S not R.) These functions are intended primarily as a way to confirm that simulations and estimations with simulated data work in the same way in both S and R, not as an improved RNG. (It has only been tested in Splus 3.3) Default and other RNGs can still be used and are preferred for both speed and theoretical reasons.

Examples

# NOT RUN {
setRNG(kind="Wichmann-Hill", seed=c(979,1479,1542), normal.kind="Box-Muller")
rnorm(10)

sim <-function(rng=NULL)
 {if(!require("setRNG")) stop("This function requires the setRNG package.")
  if(is.null(rng)) rng <- setRNG() # returns setting so don't skip if NULL
  else        {old.rng <- setRNG(rng);  on.exit(setRNG(old.rng))  }
  x <- list(numbers=rnorm(10))
  x$rng <- rng
  x
 }

z <- sim()
sim()$numbers
sim(rng=getRNG(z))$numbers
z$numbers

# }