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

[Jean-Pierre Vesinet <jpvesinet@xxxxxxxxxx>]

Erik van der Werf wrote:

> The problem is not that I don't understand the rules. The problem is
> that I don't want to check the future of each position stored in the
> transposition table with the history of the current position where a
> transposition (TT) is found. 
> 
> Think of the following line:
> A-B-...-C-...-B-...
> 
> since B is illegal because of superko this will become:
> A-B-...-C-...-D-...
> 
> C is stored in the TT 
> 
> Now after some time we search:
> E-F-...-C-...
> 
> C is found in the TT so this line will not be expanded further.
> 
> However the value of C is based on the continuation C-...-D where
> C-...-B-... may give a higher score (unless somebody could prove to me
> that this cannot exist in real Go).
> 
> I hope this clarifies the problem.

It does, thanks. You are using superko with a TT. This is a source of
possible errors. Since a situation's value is defined by the entire move
history, there is no easy way to ensure that a stored value in the TT can be
reused. There was a "transpositions" thread about this last year
(http://hamelin-zm.nom.fr/go-computer/mail10.html). No one had a fail-safe
solution IIRC.

My program uses only the basic ko rule. It equates a long cycle to jigo, so
each board situation (= position + next player + ko forbidden point) can be
stored easily in the TT. Basic ko is less satisfactory than superko, because
for example optimum play may depend on the value of komi, but at least the
values in the TT are well defined.

Jean-Pierre