On Sat, 2007-01-06 at 12:26 +0000, Jacques Basaldúa wrote: > Hi > > Today computer tournaments should be played under Chinese > as, I think, most agree. This is a thread about how could > Japanese rules be implemented.
I don't think this is that complicated, but I haven't tried to implement it yet either. The fact is, even my dumbest programs have a strong sense of what is dead and what is alive. I once experimented with this and have the code in my program that displayed a special board that shows me all the dame and unclear groups and it was quite accurate (at least to the level of my own understanding.) It basically shows a board that is filled with white and black stones unless the point is in question. I am quite sure that if I tried to implement Japanese rules it would not be difficult although David Doshay and others seem to think this phase has something to do with exceptional "skill that should be rewarded" - it's not. It's just another rule-set albeit an inferior one. I'm not saying that it's easy to get it right in every case. However, unless I go commercial I have no interest in wasting my time implementing what I consider a broken and inferior rules-set. It doesn't make me feel that my program is exceptional - it would still be a weak program. So I see no point in moving towards Japanese rules for computer/computer play. It just doesn't make any sense whatsoever to open that can of worms for no good reason - unless you just like the idea that humans will occasionally have to get involved in correcting the score and the inevitable problems this will cause. (And I would point out that on occasion it causes problems even with human play.) Wouldn't it be much finer to focus our efforts on making the program play better? - Don > Open question 1: I think dame has to be filled. Programs like > gnugo, even old versions, can to that in fractions of a second > per move. Filling dame requires almost no extra time and avoids > silly counting errors in things like: > > · · · # O · > · # # # O · > · # · O(·)· where (·) could be miscounted. > > Open question 2: A public domain library should determine > what is alive and what is not. Even at the price of some > "missimplification". Today, programs like gnugo do that > well enough, perhaps only some sekis are misunderstood. > A group would be alive or dead under CGTR200x > (Computer Go Tournament Rules or whatever) and that would > be final and unambiguously determine the score. All programmers > would use the same public domain library as the tournament > directing program, so they would all agree. > > Epilogue: If that was done some day, of course, there > would be people finding super complicated positions > for which the eval fails. But that doesn't change > anything and there is no need to rush with a new > version. Only every five years or so, after long > pondering, the next specification of the tournament > rules should come out. Any sport has (by orders > of magnitude) more referee errors than go and that > does not invalidate sport. The probability of recreating > a known error in a real game is close to zero if the > eval function is sound enough. > > > Jacques. > _______________________________________________ > computer-go mailing list > [email protected] > http://www.computer-go.org/mailman/listinfo/computer-go/ _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
