Posted by Rufus on 10/03
I know people have brought up on channels and in discussions
lately doubting the randomness of our number generator. I did
12 separate tests on our number generator, each with somewhere
in the neighborhood of 300 million iterations and calls to
random number function. 8 of those were straight tests, 4
of them bell curve tests.
The straight number test came up with expected results, the
numbers were equally distributed with a variance of less than
0.001 percent. Some of that error can be associated with the
natural rounding of C in the analysis numbers.
The 4 bell-curve tests were happy and bell-like, with an
average variance of about 0.001 percent as well. Again,
the variance is likely smaller, but I needed to have numbers
small enough to read (300 million is a lot of digits!).
Anyway, I took some time to read up on random number theory,
and our method of doing randoms is reasonably primitive. However
from what I could dig up, the difference between the two most
common methods (higher order and lower order bit randoms) is
almost negligible on current generation chipsets.
I am planing to do some pattern testing, and Ea! has agreed to
look into some alternatives to our current methodology of
generating random numbers, but so far our testing shows
a reasonable amount of 'randomness' in our random number
generator.
'Bad runs' can likely be chalked up to 'karma' but I'm wondering
if the rules of service apply (uh oh! Analogy time!):
On the average, if you get bad customer service at a you will remember that incident and
talk about it for the next 23 years. If you receive good
customer service at you
will likely only remember it for 18-24 days.
Maybe a stretch on the analogy, but if things are going right,
many many many times in a row, it's harder to notice =)
Food for thought (and yes, running that many iterations takes
a while, I'm a glutton for punishment).
-Ruf
From: Christopher
Sunday, October 01 2000, 04:13AM
Rufus, get a job.
From: Rufus
Sunday, October 01 2000, 01:15PM
I'd like to, but until then, you'll just have to deal with me
working on Legend, testing stuff, listening to people and
acting on it.
thanks for the encouragement.
-Ruf
From: Splat
Sunday, October 01 2000, 11:54PM
off back to your Karaoke bar Christopher.
From: Christopher
Tuesday, October 03 2000, 02:04AM
I didn't mean it like that :P

|