Re: random .. ?

From: David Wagner (daw_at_taverner.cs.berkeley.edu)
Date: 02/01/05 

Date: Tue, 1 Feb 2005 22:33:12 +0000 (UTC)

infobahn wrote:
>There's a rather unwieldy series of tests collectively known as
>DIEHARD, which the cognoscenti claim[1] can give you good information
>on how statistically "random" your series of numbers appears.

DIEHARD is ok, I guess. If your sequence fails DIEHARD, that's pretty
good evidence that it is not random. But if your sequence passes DIEHARD,
about all you can conclude is that it might be ok. It is very easy
to fool DIEHARD into thinking that a non-random sequence is random.
For instance, the sequence MD5(0), MD5(1), MD5(2), ... fools DIEHARD.
The good news is that DIEHARD seems better-suited for testing outputs
from some physical true-random process than for testing algorithms.