Re: [gnugo-devel] (no subject)

[Arend Bayer]

Hi,

> Hello, I am going to begin work on my honors thesis for university and was 
> thinking to "Develop software to encourage and facilitate the advancement of 
> Computer Go."  I was wondering what everyone thinks that Computer Go needs in 
> order to advance.  (By advance I mean for computer Go programs to become 
> stronger.)

As far as GNU Go is concerned: Most critical are probably fuseki, better
strategic understanding (particularly of weak groups) and better
life-and-death analysis. For Computer Go as a whole you should better
ask on the computer-go list.

> Some ideas that I had came up were tools to automate testing and strength 
> evaluation of a Go program. (Just like GnuGo's testing suite but at the end 
> get a score based on EndGame, Opening, Life/Death, etc.  To quickly diagnose 
> where a program could use improvement.)  Also, I think maybe some 
> foundational algorithms to determine life/death of a group, score a game, and 
> perhaps endgame analysis would be nice.

Systematic testing is difficult as far as I can see (apart maybe from
endgame, where probably the "endgame contest" feature of some of the
twogtp.* scripts in the interface/gtp_examples directory of gnugo should
already give a pretty good measurement). What ideas are you talking of?

Arend