Go

[Volker Pittlik]

...Be sure : no privacy infringement from my side here Volker...

OK.

...PS Remi Coulom answered with a much more complete and up-to-date list of top computer go references.

To late, I'm already infested! Warning: Don't click here! and delete all links at CCC! Here is just another post I read with interest. There seems to be step forward in go programming too.

Regards

Volker

[Daniel Shawul]

Well I clicked
I roughly went through the paper and what? no alpha-beta?
no evaluation? This is clearly chaos:):)
But I like it since it makes you look for your own ways, and dig
deep in to other algorithms. Don't have much too do with comp chess lately so I might try this to keep my mind busy!
It looks a backward step to go back to Checkers, so why not Go or Shogi. Is shogi as complicated too? (Tord?)

[Volker Pittlik]

Well I clicked ...

Sorry I split this topic from Marc's posting because I think it is to interesting and does not have much to do with Marc's topic.

I already downloaded GNUgo and qGo as GUI and won (!) a game against it on a 9*9 board. The only problem is that I don't know how I managed it .

Volker

[Uri Blass]

Well I clicked
I roughly went through the paper and what? no alpha-beta?
no evaluation? This is clearly chaos:):)
But I like it since it makes you look for your own ways, and dig
deep in to other algorithms. Don't have much too do with comp chess lately so I might try this to keep my mind busy!
It looks a backward step to go back to Checkers, so why not Go or Shogi. Is shogi as complicated too? (Tord?)

I read and I read things that are simply illogical.

"The fastest programs can
assess just 50 positions a second, compared with 500,000 in
chess."

This seems to be nonsense.

Maybe the best programs evaluate just 50 positions per second but the fastest programs should evaluate more positions.

The first evaluation that I can imagine is simply 0 if the game is not finished and using that type of evaluation a program can evaluate clearly more than 50 positions per second.

Uri

[Daniel Shawul]

I think they meant to compare the branching factor. Ofcourse if you have a win/loss/draw only eval the nps is going to rise for both, because of more alpha-beta cuts. Go has a braching factor (unpruned) of 200 as compared to 35 of chess's. Now with the exponential increase ,I guess that the number should be right for the depth achieved at tournament time control.
Daniel

[Vladimir Medvedev]

I think they meant to compare the branching factor. Ofcourse if you have a win/loss/draw only eval the nps is going to rise for both, because of more alpha-beta cuts.

1. It is VERY NON-TRIVIAL task to build even winn/loss/draw eval in Go (especially for Japanese rules). The most difficult thing is to detect terminal position, when it is safe to say PASS instead of protecting your own area or invading into opponent's one.

2. I'm 2-3 kyu in Go (it can be compared to chess rating 1800-1900) - and being only an average amateur, I still can give 9 stones handicap to world champion Go program... And win by a great margin...

[Daniel Shawul]

I don't know why they have to make the board so large. If chess was played on a 19x19 board with some more pieces then it might also be very difficult for computers to compete with humans?
From the way you demolished the computer, Comp Go looks very primitive at 19x19 board.
How does computers perform on a 9x9 board? At which board size are computers assumed to be competitive?

[Uri Blass]

I think they meant to compare the branching factor. Ofcourse if you have a win/loss/draw only eval the nps is going to rise for both, because of more alpha-beta cuts.

1. It is VERY NON-TRIVIAL task to build even winn/loss/draw eval in Go (especially for Japanese rules). The most difficult thing is to detect terminal position, when it is safe to say PASS instead of protecting your own area or invading into opponent's one.

You can consider the position as terminal only after both sides said pass.

If you can continue the game and make a blunder by not saying PASS then you can consider the position as not terminal position.

Of course good evaluation may evaluate it as terminal but when I think about fast program I think about the most simple evaluation.

Uri