On Wed, Jan 23, 2013 at 5:26 AM, Johannes Meixner <jsmeix@suse.de> wrote:
On Jan 22 13:26 Claudio Freire wrote (excerpt):
If you want to test randomness, you can use a better test suite (like NIST's, one that won't result in false positives). If you're already using the better test suite, then the false positive is not a false positive, it's a positive, it shows some cases when the generator isn't random, and should be investigated.
Only out of curiosity:
A random number generator that generates only the numbers 0 and 1 should produce a sequence of n same numbers with probability 2 * ( 1 / ( 2^n ) ) = 1 / 2^(n-1) = 2^(1-n)
There is a non-zero probability to get a sequence of same numbers with arbitrary length n.
Because of the randomness a sequence of n same numbers can happen at any time.
I wonder if it is possible to make a test suite that finishes in finite time that never shows false positives or false negatives.
I think it is not possible to make such a test suite.
It's far more than that. NIST also tests conditional probabilities, distribution independence over subsampling, and a lot of important tests if one is to use that generator in security-critical applications. In any case, all those tests are probabilistic, and can be parameterized to decide how often you're willing to get false positives. If you set that to "2^(-64)", you'll be a very old man by the time it fails. Or not and be SOL. But in any case, it wouldn't by any chance be a common occurrence, and you would be safe to consider it a true bug worthy of a bug report if it failed several times in short succession. -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org