[Pachi] [Computer-go] MCTS and perfect endgame

Jean-loup Gailly jloup at gailly.net
Sun Jul 3 17:26:56 CEST 2011


Leon,

> return (10 - rand() % 20) + board->komi + board->handicap +
scores[S_WHITE] - scores[S_BLACK];

yes I had misunderstood your proposal, thanks for clarifying it. It is an
interesting
idea. Would you keep the constants (10 and 20 in your example) really
constant
throughout the game, or vary them depending on win ratio and how far we are
in the game and if so how?

> I understand that it is imposible to believe but it works and i know why.

Many things in MCTS are counter-intuitive so I am quite willing to believe
that
it can work. But I would like to know your explanation of "why". It is easy
to try
winning by a larger points margin. The difficulty is to avoid losing a won
game because
of this, and I am curious how your proposal would do this.

Jean-loup

>
> Sorry, You totally misunderstood.
>
> RANDOM means random for each final position.
>
> EXAMPLE:
> ---------
> board.c
> 1406>    return (10 - rand() % 20) + board->komi + board->handicap +
> scores[S_WHITE] - scores[S_BLACK];
> ---------
>
> I understand that it is imposible to believe but it works and i know why.
>
> I tested with pachi and gogui and can send you some console outpout.
>
> I thought of dynkomi also but it was at once rejected.
>
> That is what I do, thinking out of the box.
>
>
> Leon.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://rover.ms.mff.cuni.cz/pipermail/pachi/attachments/20110703/eb1ab892/attachment.html>


More information about the Pachi mailing list