[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] two stone games

From: Arend Bayer
Subject: Re: [gnugo-devel] two stone games
Date: Thu, 31 Jul 2003 19:16:33 +0200 (CEST)

Stéphane wrote:

> I wonder if we could use an alphabeta-like idea to speed up that code.
> The idea is that, fondamentaly, it works by reducing the opponent's
> territory step by step, so this is the first place in gnugo where we
> explore a reading reading while having have a clear valuation (or
> should we call it a pay-off?) along the paths.
> So it would be really nice and faster if we could prune some variations
> with alphabeta ideas, like : "ah, I have already proven that I can
> reduce that territory by 10 points in that old variation, but now
> my opponent claims it can prevent me from getting more than 4 here,
> so I can cut some work here"

This sounds good but does not fit to how the code currently works. At
the moment, the recursive breakin reading can only decide _whether_ there
is a break-in possible, not how deep it will be. To decide how deep, the
recursive code is, in case of success, called again with a shrunk goal

Your idea seems worthwile, not so much for performance but for precision
to avoid the crude heuristic mentioned in the previous sentence. Most of
the time the break-in code seems to be investigating unsuccessful
break-in attempts, and there alpha-beta reading wouldn't make a


reply via email to

[Prev in Thread] Current Thread [Next in Thread]