[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [computer-go] Is seki problem solved ?

My program Viking has a life and death reader which is designed (in
theory) to always provide correct answers given enough time to read out
the situation (enough time = infinite time...).

For seki this means that if some black stones provably cannot be
captured by white they have to be alive. If the reason for this is
because they have two eyes or seki does not matter. 

The problem with this approach that my program only plays correctly when
the seki is easy to read. In the middle game group tactics are often
(always) too complicated to read out completely, but sometimes it
evaluates seki correctly in the end of the game.

A second approach is to implement static algorithms that detects seki
without search. If this is done efficiently then no search is necessary.

The problem of this approach is that you will only be able to find
certain classes of seki if your algorithm is not general enough. There
are many rare kinds of seki that are easy to miss with static analysis.

(I am also now ignoring the fact that all groups surrounding the stones
in seki has to be unconditionally alive and in many situations occurring
in messy games the tactical problem should often be seen as semeais with
more than two groups until proven otherwise)

In summary allowing pass for the defender in a tactical search that
tries to prove if capture is possible do not need code for seki. Such
search would probably benefit greatly if seki can be evaluated
statically if can program it. But if you really want to know that there
is a seki because some other part of the program need to know exactly
why a group is alive then this might not be enough.

--
Magnus Persson
Center for Adaptive Behavior and Cognition
Tel: +49-(30)-82406-350
Cell phone: +49 163 6639868

> -----Original Message-----
> > Is seki problem solved ?
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/