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

Re: computer-go: Perl Module for next move.



On Tue, Jun 05, 2001 at 11:19:03PM +0200, Grajdeanu, Adrian wrote:

> Heikki, I don't need gnugo to enumerate all valid moves. I've already got
> that covered. Though one thing I missed was what you call a super-ko. I
> learned of it through experience and actually wanted to ask some go
> authority. So to make sure I understand it right, the go rule sais that no
> position may be repeated, right?

This is one point where the rule sets differ. The main idea is to prevent
endless recapturing loops. Some rules ("japanese") require that before
recapturing a ko, you need to play somewhere else. Other rules ("super-ko")
prohibit the creation of a position that has already occurred in the game.
Others can get more complex. For computer go, the super-ko rule is easy to
implement, and solves some pathological cases like triple kos and
multi-stone repeating sequences.
 
> What happeds if GnuGo is asked to score a position where human players
> don't see the end? Is the resulting score a decent one? For if it is, then
> maybe I can use it.

It will give you a fairly good score, but not a perfect one. It can even
make mistakes in scoring finished games, but probably so seldom as not to
disturb your NN experiments.
 
> To be honest, I'm skeptical too. I just wish to run an experiment, that's
> all.

Like I said, best of luck. And keep us posted on your progress!

-Heikki

-- 
Heikki Levanto  LSD - Levanto Software Development   <heikki@xxxxxxxxxxxxxxxxx>