Pseudorandom numbers not so random

I have a bit of a bug in our game that's upsetting my players quite a bit. It's the dice rolls.

I know. Java (or some library its using) makes it's pseudorandom numbers and on average they're - well, random. But we've noticed some rolls that just don't *look* quite as random as we would expect. Such as (rolling several d20 successively)

1,1,3,1,2,3,2,1,20,4,20,1,19,17,19,9,5,2,1,2,1,1,13,20,13,15,18,14,18...

along those lines - 1s or 2s or 3s will sometimes repeat (which is fine, that happens irl), but they'll happen all grouped up - so with the law of large numbers, everything looks good but in the short term things are odd. Does that make sense? Imagine flipping a coin and getting 10 heads in a row, then 10 tails in a row. The ratios you got were 50/50, but that's not really what we expect when we say random, right? Then well also see the ABA number pattern thing, like the (19,17,19). Which again, happens irl, but.. more often in d20pro?

Yeah, it's a weird thing and I'm sure it's java, but, curious if there's *any* way I'd be able to affect it some. My players don't mind the 20 rolls so much (unless I'm making them), but they get annoyed rolling 4 or 5 1s in a row In any case, even if I can't fix it I'd like to know if there's some magic step I can try that might convince my players it's *fixed*. Outside of reinstalling my OS (unless I really need to).

Also, incidentially, I think there are 'true random' number generators in java (extra libraries) that might be better than the built-ins? Thanks.

 

My players don't really care so much about the statistics as what they're repeatedly experiencing. If I were to roll ten 20's in a row, that would not be statistically insignificant. (In fact, the 'real' probability of getting this is 9.765625 × 10-14) That's a small sample size, and with just 10 rolls the odds of that happening are so low we don't have names for the decimal places in discussion.

ogexam: Thanks for the good example, it proves what I'm saying about the numbers - the *total average* is good (pseudorandom number generators typically guarantee this), but look at the number of 10's vs. the number of 18s. You should not be getting a bell curve here, but a generally straight line. On any large scale (500 should be pretty good), you really, really shouldn't be getting twice the number of rolls of one dice than another - your expected value here is 25, and you're getting 38s and 15s.

Which is bad, because my initial assumption was that the pseduorandom number generator java was implementing on my computer was bugged by my computer somehow. Your results make me think it's worse than that, I'm pretty sure java's included random library doesn't pass all the Marsaglia diehard tests.

Regardless of all the above, there are better java pseudorandom number generators than the included libraries, such as javax.crypto.SecureRandom, which I think is slow.

'Uncommon Maths' is supposed to have a good library? But I don't know what it is.

Also, apparently someone's implemented Mersenne twister, http://www.cs.gmu.edu/~sean/research/

This also looks like it could be a simple quick patch to slightly improve the numbers: http://www.javamex.com/tutorials/random ... hift.shtml This little trick would probably at least calm my players qualms.

edited: removed the trolling flamey stuff

 

Sweet. I think theres a similar one here: http://www.math.sci.hiroshima-u.ac.jp/~ ... T-java.zip

That one is free to use for commercial http://www.math.sci.hiroshima-u.ac.jp/~ ... ICENSE.txt, it kind of looks like it's a newer file - 2007 instead of 2004, but maybe it's not as easy to implement into the code, I dunno. Also, it's supposed to be a lot faster, at least the algorithm. The 2004 one seems to implement a lot of hacks to account for java limitations, but maybe thats a good thing, I dunno. Well hope that helps some.

Also, d20pro is tops. No other software I've tried comes close to d20pros usability and features and such. Just thought I'd point that out.