(I just forgot to mention that we take care of rotations / symetries)
I was puzzeling what is the best way to organise an opening book?
>>
>
> Our solution was the following. We have two possible structures:
> 1) a big set of SGF games;
> 2) an opening book in which one line contains p,w,l where
> * p is a situation
> * w is the number of wins
> * l is the number of losses
>
> We can convert 1) to 2) easily.
>
> The advantages are:
> 1) is quite comfortable for interactions with humans or will self-play
> games;
> 2) is much faster to read and to apply for choosing a move with various
> formulas (it's quite analog to a MCTS tree).
>
> In particular, self-play for small boards is efficient for building good
> opening books - well, I think it's better if you can have human experts, but
> if you spend sufficiently many computational power on it you can build a
> good opening book by self-play (the algorithm is by the way quite related to
> MCTS).
>
> Best regards, hope it helps,
> Olivier
>
>
>
--
=========================================================
Olivier Teytaud (TAO-inria) olivier.teyt...@inria.fr
Tel (33)169154231 / Fax (33)169156586
Equipe TAO (Inria-Futurs), LRI, UMR 8623(CNRS - Universite Paris-Sud),
bat 490 Universite Paris-Sud 91405 Orsay Cedex France
(one of the 56.5 % of french who did not vote for Sarkozy in 2007)
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
