Re: [gnugo-devel] look at regress file 13x13b:4

[Evan Berggren Daniel]

On Sun, 30 Mar 2003, kevin yong wrote:
> So, problem is: in general, pattern match is
> local/small, it is mis-used in a big environment.
> Specifically in the case of 13x13b:4, it is currently
> at endgame stage now. This Joseki pattern to
> expand-territory and moyo should only applicable at
> start-middlge stages of a game, provided that there is
> enough room to expand. This Joseki pattern match over
> looked the later.
>
> Suggested solution: there are several options
> available (at least):
> 1.
> - before shapes() analysis called, insert another
> step, let?s temporary named it ?stage_judgement()?,
> which will give a general judgement it?s at
> open/middle/end game stage;
> - inside shapes(), after a expand territory/moyo
> pattern found, should also look at what stage of game
> we are, then decide if the pattern is applicable or
> not.
> 2.
> - inside shapes(),after a expand territory/moyo
> pattern found, add a step to check if there is enough
> room to expand before add big points for this pattern.

These are certainly all possible.  However, my guess is that it will be
easier for gnugo to (eventually) determine the size of such moves on its
own, than to determine whether or not a pattern is correct in the general
case.  I think the most obvious and immediate problem is that the patterns
ED66 and ED67 are at the very least overvalued.  The constraints also need
improving, in all likelihood.  I think fixing these patterns is the
solution Gunnar had in mind when he suggests 13x13b:8.

Evan Daniel