Talk:Diehard tests

From Wikipedia, the free encyclopedia

You have new messages (last change).

I've based this largely on the descriptions in Marsaglia's C code, which I don't understand too well. I would be obliged if some statistician would rewrite the descriptions more accurately. Tualha 4 July 2005 12:22 (UTC)

"Collectively they are considered one of the most stringent such tests known". This is sadly no longer true. They are good tests and still worth doing, but there are tougher ones available now. The main problem is that Diehard tests only about 11megabytes of data. As some modern applications use as much as several terabytes of random data, Diehard might not be able to detect large-scale patterns that are important to some people. Just off the top of my head, more stringent tests include NIST, SPRNG, L'Ecuyer, gjrand, and probably heaps of others. This is not supposed to belittle Marsaglia. Diehard is a big improvement on what was easily available before, and the newer tougher tests are all to some extent inspired by Diehard. 203.164.223.250 01:42, 17 April 2006 (UTC)

[edit] Dead link?

The original link seems to be dead. The mirror link works just fine, though.

Retrieved from "http://en.wikipedia.org../../../d/i/e/Talk%7EDiehard_tests_93ff.html"