Re: [computer-go] Computer Olympiad photo

[tesuji@xxxxxxxxx]

Some of the discussions here seem to belong in a computer-chess mailing
list rather than here.

Concerning using brute-force, it's not so unreasonable for 9x9. Generally
games for 9x9 are one of two types:
- A whole-board fight from start to finish
- The board gets divided up fairly quickly in stable parts and it the
whole game is like a big end-game.

In my experience, games played by software tend to end up in the latter
category much more often than the first. So a reasonable but fast
evaluation combined with brute-force search stands a good chance against
exisitng software which is designed for 19x19.

But in games of the first category, just brute-force will stand no chance
whatsoever. Against any 1-dan player making sure the game develops into a
game of the first category, a brute-force program won't stand a ghost of a
chance.

However, end-game is much more important in a 9x9 game than on bigger
boards, even in a hectic game. So a program that can do that well will
always have a good chance against software that was made for 19x19 but
which is playing on a small board. But only up to a certain level...

In my opinion 9x9 is such a different game that if as much effort was put
into it as for 19x19 a reasonable level (1-dan?) could have been reached.
Probably such programs would be a combination of a good evaluation and
whole-board search. If you can make a program that can read 10 or 12 ply
deep (or even more) and still have a reasonable idea about territories,
eyes, etc. it should do much, much better than any existing program.

    Mark Boon

> Vincent Diepeveen wrote:
>> I feel some commercial go programmers should enter his
>> program in the 9x9 ICGA go olympiads to let the chess programmers feel
>> where their place is in history... ...in the far past
>
> "this uni guy with a programming technique from some decades ago"
> already defeated Go4++ once last year and had quite reasonable results
> in this years tournament. I think you're too optimistic about the
> strength of the commercial Go programs. In fact, I think that if one is
> able to build a program that plays significantly stonger than the latest
> version of GnuGo, there is a very good chance to win against any
> commercial program.
>
> Erik
>
>
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://computer-go.org/mailman/listinfo/computer-go
>

_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go