Re: computer-go: Evaluating positions

[Heikki Levanto <heikki@xxxxxx>]

On Mon, Jun 18, 2001 at 09:08:32PM +0200, Vlad Dumitrescu wrote:
> I think there is a different meaning of the word "influence" that we are
> using. Otherwise I agree with you completely! :-)

Probably.

> I call influence of a group of stones a quality unaffected by the strength
> of that group. Stones of opposite color may be removed from the board when
> estimating it. Each empty point has a value between 0.0 and 1.0.

Sorry, I do not understand where this number comes from, and what does it
mean?
 
> The strength is also measured from 0.0 to 1.0, where 1.0 is
> unconditionally alive (even if passing all the way).

Again, could you elaborate? What is the value for a group that can stay
alive if every attacking move can be answered by a suitable defence? (that
is, in ordinary go terms, "unconditionally alive"). What does value 0 mean
(no group at all?). 

> Does that seem too strange?

Strange, yes. Too strange, I can not say yet. 

Some critical questions and comments:
  - You talk of a strength of a group. How do you define a group?
  - How about interactions between friendly groups?
  - I have a strong feeling that the essence of go can not be captured
    in one number, not even one number for each point on the board. 
  - If your influence function should have a good correspondance to 
    "what is really going on in the game", that should be easy to show,
    for example by computing it at every move in the game, and comparing
    to some humans explanation of what is happening... Something like
    "black got ahead in the opening, but white caught up in the middle
    game and was actually winning until the blunder at move X, which 
    lost the game".
    
Even if I remain sceptical, I wish you best of luck. Keep us informed on the
progress.

- H

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