Create account / Log in

Dice

Topics related to the main Vassal engine.

Moderators: Tim M, uckelman

Dice

Postby rhc9 » February 23rd, 2009, 2:47 pm

I am a newbie, and although I have searched the achives, I cannot find an answer to this question. I am playing Twilight Struggle and it seems to me the die is streaky to the point of being mathmatically inaccurate. I know that some would say the same about the ASL module because a twisted dice attachment is offered to replace the original dice. Is this really a problem? And if so, can it be fixed?
rhc9
 
Posts: 1
Joined: February 23rd, 2009, 2:38 pm

Dice

Postby uckelman » February 23rd, 2009, 3:26 pm

Thus spake "rhc9":
I am a newbie, and although I have searched the achives, I cannot find an ans
wer to this question. I am playing Twilight Struggle and it seems to me the d
ie is streaky to the point of being mathmatically inaccurate. I know that som
e would say the same about the ASL module because a twisted dice attachment i
s offered to replace the original dice. Is this really a problem? And if so,
can it be fixed?

No, it is not really a problem. The standard dice roller uses a random
number generator which scores well on a variety of randomness tests.* The
problem is human perception, not the dice roller. People tend to
dramatically underestimate the probability of hot or cold streaks, and
anyway one game of Twilight Struggle contains too few die rolls to give you
a statistically significant sample.

(Also, the Mersene Twister roller you refer to has been the default VASSAL
roller for quite some time now, at least since 2.9.something.)

To date, I have seen no statistical evidence that there is anything wrong
with the current dice roller, and when assessing randomness, statistical
evidence is the only kind of evidence which counts. If someone can show
me a large sample generated by our dice roller which fails a reliable
battery of statistical tests, I'll reconsider, but I'm not holding my
breath.



* I've checked this in the past and posted results someplace, possibly the
vaslmapcabal list or our old Yahoo groups list. Someone who claimed that
the roller was bad supplied me with a corpus of ASL logs containing
thousands of die rolls. I extracted these from the logs, and then ran Ent on
them (http://www.formilab.ch/random) and got the results you'd expect from a
good (non-cryptogrpahic quality) pseudorandom number generator. If you ran
a tougher battery of test (e.g., DieHarder) on these results, I'm sure that
you could tell that they came from a pseudorandom source, but the differences
are going to be miniscule, on the order of a few die rolls over the course of
your whole gaming lifetime.

--
J.

_______________________________________________
Messages mailing list
Messages@forums.vassalengine.org
http://forums.vassalengine.org/mailman/ ... engine.org

Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
uckelman
Site Admin
 
Posts: 8373
Joined: December 10th, 2007, 9:48 am
Location: Durham, England

Dice

Postby mkiefte » February 23rd, 2009, 3:36 pm

There's nothing in the VASSAL code that would suggest a problem with the die roller. The die roller in VASSAL uses the same random number generator as many other Java applications and it would surprise me if there was a noticeable problem. If you were really concerned, you could post a saved game log and a stats geek will calculate the probability of the rolls in question (keeping in mind that there are probably thousands of rolls being generated each day).

Just to put this in perspective, the odds of getting the same roll three times in a row in a 1d6 is 2.8%. In a game, that's probably going to happen a couple of times. The odds of rolling the same value on a 2d6 three times in a row is 1.4%.

- M.

2009/2/23 rhc9 <messages@forums.vassalengine.org (messages@forums.vassalengine.org)>
I am a newbie, and although I have searched the achives, I cannot find an answer to this question. I am playing Twilight Struggle and it seems to me the die is streaky to the point of being mathmatically inaccurate. I know that some would say the same about the ASL module because a twisted dice attachment is offered to replace the original dice. Is this really a problem? And if so, can it be fixed?








Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
mkiefte
 
Posts: 1144
Joined: January 5th, 2008, 1:29 am
Location: Halifax, Nova Scotia, Canada

Dice

Postby Brent Easton » February 23rd, 2009, 8:12 pm

On 23/02/2009 at 11:36 AM Michael Kiefte wrote:
There's nothing in the VASSAL code that would suggest a problem with the
die roller. The die roller in VASSAL uses the same random number
generator as many other Java applications and it would surprise me if
there was a noticeable problem.

Due to the unending complaints from the players of one particular module, we replaced the standard Java Die Roller with a Mersenne Twister algorithm. We no longer use the 'standard' Java random nuumber generator.

B.


_______________________________________________
Messages mailing list
Messages@forums.vassalengine.org
http://forums.vassalengine.org/mailman/ ... engine.org

Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
Brent Easton
 
Posts: 2863
Joined: December 21st, 2007, 3:06 am
Location: Berry, NSW, Australia

Postby TOPO » February 23rd, 2009, 8:42 pm

Sorry for this horror vacui reply but... this seems like a serious case of Murphy's Laws on action... :roll:
A dice (electronic or real) always goes wrong (vs you)... :P
User avatar
TOPO
 
Posts: 42
Joined: February 10th, 2009, 3:58 pm

Dice

Postby mkiefte » February 23rd, 2009, 8:47 pm

In fact, I would expect the dice that are provided in most games to fail many trivial randomness tests.

Dice are not perfect cubes and if you took a digital calliper to them, you'll find huge deviations. In addition, most game dice have hollowed-out pits that favour rolls of 6 because the other side (1) is typically much heavier. My prediction is that the java.util.Random class is better than that. I'd be interested to see evidence to the contrary however.

Here's an interesting link:

http://www.maa.org/mathland/mathtrek_10_26_98.html

- M.

2009/2/23 Joel Uckelman <uckelman@nomic.net (uckelman@nomic.net)>
Thus spake "rhc9":
I am a newbie, and although I have searched the achives, I cannot find an ans
wer to this question. I am playing Twilight Struggle and it seems to me the d
ie is streaky to the point of being mathmatically inaccurate. I know that som
e would say the same about the ASL module because a twisted dice attachment i
s offered to replace the original dice. Is this really a problem? And if so,
can it be fixed?


No, it is not really a problem. The standard dice roller uses a random
number generator which scores well on a variety of randomness tests.* The
problem is human perception, not the dice roller. People tend to
dramatically underestimate the probability of hot or cold streaks, and
anyway one game of Twilight Struggle contains too few die rolls to give you
a statistically significant sample.

(Also, the Mersene Twister roller you refer to has been the default VASSAL
roller for quite some time now, at least since 2.9.something.)

To date, I have seen no statistical evidence that there is anything wrong
with the current dice roller, and when assessing randomness, statistical
evidence is the only kind of evidence which counts. If someone can show
me a large sample generated by our dice roller which fails a reliable
battery of statistical tests, I'll reconsider, but I'm not holding my
breath.



* I've checked this in the past and posted results someplace, possibly the
vaslmapcabal list or our old Yahoo groups list. Someone who claimed that
the roller was bad supplied me with a corpus of ASL logs containing
thousands of die rolls. I extracted these from the logs, and then ran Ent on
them (http://www.formilab.ch/random) and got the results you'd expect from a
good (non-cryptogrpahic quality) pseudorandom number generator. If you ran
a tougher battery of test (e.g., DieHarder) on these results, I'm sure that
you could tell that they came from a pseudorandom source, but the differences
are going to be miniscule, on the order of a few die rolls over the course of
your whole gaming lifetime.

--
J.





Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
mkiefte
 
Posts: 1144
Joined: January 5th, 2008, 1:29 am
Location: Halifax, Nova Scotia, Canada

Dice

Postby uckelman » February 23rd, 2009, 9:07 pm

Thus spake Michael Kiefte:
In fact, I would expect the dice that are provided in most games to fail
many trivial randomness tests.

Dice are not perfect cubes and if you took a digital calliper to them,
you'll find huge deviations. In addition, most game dice have hollowed-out
pits that favour rolls of 6 because the other side (1) is typically much
heavier. My prediction is that the java.util.Random class is better than
that. I'd be interested to see evidence to the contrary however.

Here's an interesting link:

http://www.maa.org/mathland/mathtrek_10_26_98.html

The average d6 roll we were producing when I tested was vanishingly near
to 3.5. I'm sure that every one of the d6s I own is biased due to being
cheap pieces of plastic.

--
J.

_______________________________________________
Messages mailing list
Messages@forums.vassalengine.org
http://forums.vassalengine.org/mailman/ ... engine.org

Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
uckelman
Site Admin
 
Posts: 8373
Joined: December 10th, 2007, 9:48 am
Location: Durham, England

Dice

Postby mkiefte » February 24th, 2009, 1:06 pm

The average d6 roll we were producing when I tested was vanishingly near
to 3.5.

You mean java.util.Random, or plastic?

 
I'm sure that every one of the d6s I own is biased due to being
cheap pieces of plastic.

Use casino dice -- those people are serious about randomness.  They don't have hollowed-out pits and I think they're precision manufactured with really low tolerances.

But they probably cost $10 a piece.

- M.

Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
mkiefte
 
Posts: 1144
Joined: January 5th, 2008, 1:29 am
Location: Halifax, Nova Scotia, Canada

Dice

Postby uckelman » February 24th, 2009, 1:39 pm

Thus spake Michael Kiefte:
The average d6 roll we were producing when I tested was vanishingly near
to 3.5.


You mean java.util.Random, or plastic?


I mean the MT roller.

I'm sure that every one of the d6s I own is biased due to being
cheap pieces of plastic.


Use casino dice -- those people are serious about randomness. They don't
have hollowed-out pits and I think they're precision manufactured with
really low tolerances.

But they probably cost $10 a piece.

Actually, you can get pairs of used craps dice for pretty cheap in casino
gift shops, if I remember correctly.

--
J.

_______________________________________________
Messages mailing list
Messages@forums.vassalengine.org
http://forums.vassalengine.org/mailman/ ... engine.org

Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
uckelman
Site Admin
 
Posts: 8373
Joined: December 10th, 2007, 9:48 am
Location: Durham, England

Postby Rindis » February 25th, 2009, 12:17 am

I was pointed to an article talking about the problems with common Chessex and Games Workshop dice just yesterday:

http://www.dakkadakka.com/wiki/en/That% ... is+of+Dice

Note the irony that while one would "expect" more 6s, you really get more 1s. :D

Personally, I'll keep using my standard Chessex/GMT dice. I find the sharp edges of casino dice way too uncomfortable. (Though an apparent 13% discrepancy is higher than I would have thought...)

And neither physical dice nor dice-bots seem to be able to stand up to my ability to make absolutely wacky results come up more often than should be possible. :twisted:
User avatar
Rindis
 
Posts: 198
Joined: December 22nd, 2007, 12:49 am

Dice

Postby mkiefte » February 25th, 2009, 1:07 pm

2009/2/24 Rindis <messages@forums.vassalengine.org (messages@forums.vassalengine.org)>
I was pointed to an article talking about the problems with common Chessex and Games Workshop dice just yesterday:

http://www.dakkadakka.com/wiki/en/That%27s+How+I+Roll+-+A+Scientific+Analysis+of+Dice

Note the irony that while one would "expect" more 6s, you really get more 1s. :D

Wacky. 19% 1s is pretty sad.


Personally, I'll keep using my standard Chessex/GMT dice. I find the sharp edges of casino dice way too uncomfortable. (Though an apparent 13% discrepancy is higher than I would have thought...)

I like to sharpen my steel dice with a leather belt.



And neither physical dice nor dice-bots seem to be able to stand up to my ability to make absolutely wacky results come up more often than should be possible.  [Twisted Evil]

There's some confirmatory bias here.  You don't notice when the dice are average.  You only remember the times when your die-rolls either really suck or are fantastic.

- M.

Post generated using Mail2Forum (http://www.mail2forum.com)
User avatar
mkiefte
 
Posts: 1144
Joined: January 5th, 2008, 1:29 am
Location: Halifax, Nova Scotia, Canada


Return to General Discussion

Who is online

Users browsing this forum: No registered users and 4 guests