Re: [gnugo-devel] miscellanous items re gnugo

[Gunnar Farnebäck]

Michael Olds wrote:
> I am guessing that you have seen the game attached before, but just in case!
> I am a very weak player (used to play some 40 years ago, just took it up
> again), but I lost in this game by only the komi (I am interested in playing
> by the old Chinese rules requiring open eyes and counting placed stones as
> territory; so actually I either tied or won this game) playing black with no
> handicap with the program pumped up to the maximum. Black plays the first
> stone on the center star and exactly mirrors white (i.e., the computer is
> playing itself).

See for example the discussion starting with
http://lists.gnu.org/archive/html/gnugo-devel/2004-08/msg00007.html

> I figured this out while trying to figure out how the program was
> "thinking". It looked to me as though there was no 'strategy', only
> recognition of situations and knowledge of how to handle them. I asked
> myself how would one use strategy against an opponent that had no strategy.

This is more or less correct, depending on how you define strategy.
GNU Go occasionally plays moves with strategic meaning, but it doesn't
do any actual planning.

> OK. Then I have an idea just to toss into the programming pool: How about
> assigning each stone 100% it's own color, with each intersection step away
> from the stone reducing that color by increments of 5%.

This is generally called an influence function in go programming. GNU
Go has one which is substantially more complex but fundamentally based
on the same idea.

> Then for each move the computer calculates the position that would
> generate the maximum (over the entire board, calculating the total
> color value of the combined stones of each color) white (or black).
> The resulting information as to the position(s) that would produce
> the maximum benefit could be used to evaluate moves generated by the
> program (closest move to maximum white is the move to make).

If this doesn't take life and death aspects into account it's useless.
Otherwise it's meaningful but not in itself sufficient for good play.
GNU Go already takes this kind of information into account.

> I would like to see the ability to play out the game to the last stone so
> that I could play the old Chinese rules properly.

The closest you can come is the --capture-all-dead option.

> All that aside, this is a very nice program...

Thanks. :-)

/Gunnar