Re: computer-go: 5x5 Go is solved?

[Erik van der Werf <E.vanderWerf@xxxxxxxxxxxxx>]

Not all positions leading to a superko have a (basic) ko, the problem is
in the continuation.

BTW even in a big cycle there can be positions that don't even have a
basic ko (e.g. one side passes or plays elswere). Under superko such
positions can still be illegal depending on how the cycle started. 
For examples see: http://www.britgo.org/rules/molasses.html

I have one solution that should work. Hashing the history. However, I'm
affraid it will cost too many transposition to be practical (still have
to test this though).

Erik.



Stefan Mertin wrote:
> 
> Erik van der Werf schrieb:
> > ...
> > I hope this clarifies the problem.
> 
> Hi Eric,
> 
> sorry, but I still donīt understand,
> why do you need to take account of all this awfully complicated
> ko - stuff in your tablebases?
> 
> Wouldnīt it be the simpliest solution
> not to store ANY ko-position at all!?
> 
> Your program then may use the super ko rule
> or even the Japanese simple ko rules,
> what you like better
> (I prefer the super ko rule for computer programs,
>  because I think there are too much ko situations,
>  that would need a very deep understanding of Go
>  if you use Japanese rules - this makes only sense for
>  very skilled human players, who really understand such situations!) ...
> 
> But tell your program,
> that a position with any ko-forbidden move
> is not a "complete position"
> and therefore not stored in your TT -
> thus when a ko comes up in a game
> the program will have to search for the next available
> "complete position" without any ko defect!
> And this shouldnīt be too much a problem, I think ?!
> 
> Also this is what every Go player would try to do
> in ko - positions: only non-understanding computer programs
> tend to loop as long as possible in ko circles,
> a real go player would ever look for a finite solution
> or (Japanese rules:) accept jigo ...
> 
> Thats just my idea of it.
> 
> Regards,
> Stefan