Re: [Bug-gnubg] Analysis/Evaluation Settings

[Holger]

At 19:16 22.01.2003 +0000, Joern Thyssen wrote:
>On Tue, Jan 21, 2003 at 06:26:53PM +0100, Holger wrote
>> At 12:21 21.01.2003 +0000, Joern Thyssen wrote:
>> >Speed option:
>> >
>> >A n-ply evaluation is an average over 21 n-1 ply evaluations.
>> >A reduced evaluation averages over a reduced set of dice, e.g., a 33%
>> >speed evaluation averages over 7 n-1 ply evaluations.
>
>Hmm, maybe I was a bit quick here. The code actually does something
>different:
>
>A n ply evaluation is an average over 21 n-1 ply evaluations. These are
>again averages over n-2 ply evaluations etc. When n-m=1 then gnubg will
>average over a reduced set of dice.

I'm not entirely sure I understood. The dice reduction applies only if it
was set? And then it only applies for the last recursion?

>Hmm, wonder which algorithm is correct/best???
>
>> Thanks for the explanation.
>> 
>> How are the dice combinations determined? Is there (can there be?) any try
>> to reduce the error introduced by this by selecting appropriate dice
>> combinations?
>
>gnubg will loop through different reduced sets of dice. The 50% speed
>option has two different sets, the 33% speed option three etc.
>
>AFAIK the sets are chosen such that "good" rolls are divided bewteen the
>different sets. For example, the double rolls.

Chosen means with knowledge which is better or just arbitrarily?

Has anybody ever done some experiments about what is better: dice reduction
or a smaller search space?

Regards,

        Holger