Dear GSL random number persons, I finally got sufficiently motivated to finish putting the rest of the diehard tests into dieharder. At this point in time there are 21 tests supported by dieharder -- 16 distinct diehard tests, two tests I invented (one of which no random number generator I've tested passes, although it is a very simple and intuitive one), a timing test, and two STS (Statistical Test Suite from NIST) tests. Any GSL rng can be tested or played with from within the tool by means of a simple command line option.
It is fairly difficult to precisely validate the resulting code. I have access to the last/best C port of diehard and have run at least the most troublesome tests in comparison with it to validate on files of rands I've generated myself. However, I've also extended the diehard tests so that there are a lot more control variables one can set at the command line, so that they can and often do use many more random number samples than diehard (which was limited by the file size reasonably accessible in 1996), and so that they consistently do a final KS test on a distribution of sample p-values with a default of 100 entries instead of just running a test once or twice and examining the resulting p's. They present in all cases but one (which is still buggy) a histogram of the p-values so you can visually inspect it for uniformity or systematic deviation. Extension always brings with it the possibility of new bugs...;-) At any rate, if any GSL-rng-using persons wish to give it a try, you can find it at: http://www.phy.duke.edu/~rgb/General/dieharder.php which should link you to a source tarball (probably the best way to play with it) or to an FC5 binary rpm if you just want to install it and run it on various GSL generators or on rands input from a file. Although it works, and works pretty well as far as I can tell from fairly extensive testing, I'm sure there are still bugs in the code. Please report any oddnesses to me if you do give it a try and it returns something unbelievable or impossible. Thanks, rgb -- Robert G. Brown http://www.phy.duke.edu/~rgb/ Duke University Dept. of Physics, Box 90305 Durham, N.C. 27708-0305 Phone: 1-919-660-2567 Fax: 919-660-2525 email:[EMAIL PROTECTED]
