[computer-go] 10k UCT bots

[Don Dailey]

I think Christoph just described the eye rule most of us are using.     
It's pretty much standard and I think it's mainly just a mechanism to 
prevent the play-outs from going on and on for hundreds of moves.      
You can probably improve on the rule,  but this is probably not going to 
make any measurable difference until you cover much more important 
things such as improving the quality of the play-outs,  what Dave Hillis 
calls "heavy play-outs."   

My suggestion is to just start with the uniform random play-out policy  
with the simple and effective eye rule we are using first,  then you can 
obsess over the eye rule and heavy play-out policy stuff later to 
improve on what you have.    

- Don



Carter Cheng wrote:
> I am curious about the eye rule situation since I am at that stage of my implementation of the board/playout code. currently I have only implemented the basic rules of the game so that no illegal moves are possible (no superko rule) but undesirable stuff like filling your own eye spaces still are. 
>
> I am curious how and when are the eye lists being calculated using the scheme described above. Also one related issue is that for certain LD shapes wont you still inadvertently fill your own eye space? 
>
> Regards,
>
> Carter
>
>
>       
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>
>