RDieHarder: An R interface to the DieHarder suite of Random Number Generator Tests

Random number generators are critically important for computational statistics. Simulation methods are becoming ever more common for estimation; Monte Carlo Markov Chain is but one approach. Also, simulation methods such as the Bootstrap have long been used in inference and are becoming a standard part of a rigorous analysis. As random number generators are at the heart of the simulation-based methods used throughout statistical computing, ‘good’ random numbers are therefore a crucial aspect of a statistical, or quantitative, computing environment. However, there are very few tools that allow us to separate ‘good’ from ‘bad’ random number generators. Based on work that started with the random package (Eddelbuettel, 2007) (which provides functions that access a non-deterministic random number generator (NDRNG) based on a physical source of randomness), we wanted to compare the particular NDRNG to the RNGs implemented in GNU R (R Development Core Team, 2007) itself, as well as to several RNGs from the GNU GSL (Galassi et al., 2007), a general-purpose scientific computing library. Such a comparison is possible with the DieHarder test suite by Brown (2007) which extends the DieHard test suite by Marsaglia. From this work, we became interested in making DieHarder directly accessible from GNU R. The RDieHarder package presented here allows such access. This paper is organized as follows. Section 2 describes the history and design of the DieHarder suite. Section 3 describes the RDieHarder package facilities, and section 4 shows some examples. Section 5 discusses current limitations and possible extensions before section 6 concludes.