RE: [Bug-gnubg] rollout settings

[Ian Shaw]

> From: max-d [mailto:address@hidden

> Can anyone give a short explanation (or links) for the 
> rollout parameters :
> 
> Full cubeless rollout without var.redn., with rotate rolls ????
> 
> I have vague idea what  variance reduction is but never heard about
> rotate rolls.
>
Rotate Rolls. The first 1296 rolls are preset to give a combination of all 
possible sequences of rolls for player X and O on the upcoming two rolls. In 
fact, looking at the change log I see:

Fri Oct  4 18:02:56 2002  Jørn Thyssen  <address@hidden>

        * rollout.c (QuasiRandomDice): rotate rolls at turn 2 and 3. 
        This should ensure dice rotating for up to 1 679 616 trials.
 
1 679 616 is 36^4, so it looks like rotation is in place for the first two 
rolls for each player.

There has been some discussion about what happens if you do a rollout of < 1296 
trials with rotation enabled. If the rolls are not random, you might skew the 
results if you do less than 1296 trials because, for example, you miss all 
sequences where X rolls 66.

There was some discussion about randomising the rolls within the 1296 sequence, 
but this has not been implemented yet as far as I know.

Until I understand what's happening, I don't trust rollouts using rotation 
unless there are at least 1296 trials.

Can anyone verify the current state of play:

What happens if trials < 1296?

What happens if trails = 36n where 1 <= n <= 36?

Are trials after 1296 random or does the rotation restart? Do we need to do 
1296n trails to avoid skew?

> when analysing a whole match ,what is the optimal rollout settings ?

AFAIK, rolling out a whole match is still not implemented. The settings exist 
in .gnubgautorc but they do nothing.

Optimal rollout settings depend on the speed of your computer. It's a trade off 
between accuracy and speed.

--Ian