Willemien wrote:
can somebody give me ideas, links and other information on how to
program (Japanese) territory scoring rules?
Read all the Japanese style rulesets here:
http://home.snafu.de/jasiek/rules.html
In particular, the Japanese 2003 Rules come the closest. For a start,
you can ignore capturable-2 and capturable-3 though because they are
needed only scarcely.
Of course, you will need to develop algorithms for partial selections of
sets of hypothetical-sequences and of hypothetical-strategies.
How to decide that the game is over.
Score the current position as if it were the final-position and compare
it to the scores of your tree-search leaves' final-positions and a
min-max on it. If they are equal, the game is "over".
How to decide what is teire (moves that costs the player points but
don't need to be played because there are still neutral unoccupied
points.
A general definition of teire is still missing. As a pretty good
approximation, implement the Japanese 2003 Rules to determine status,
eye-points and dame. Then virtually fill the dame and your program
"sees" the teire.
How to decide what is a seki and other rule beasts.
and maybe other problems related to this field.
Use the Japanese 2003 Rules, and
Types of Basic Kos
http://home.snafu.de/jasiek/ko_types.pdf
(note that I am currently correcting the definitions for active /
inactive disturbing life to refer to two-eye-formation instead of
uncapturable life)
and
External versus Internal Ko
http://home.snafu.de/jasiek/external.pdf
*************************************************************************
As you will notice very quickly, correct determination of all would be
extremely tedious. So you will need to make compromises. The ultimate
compromise (and simplification of your programming work) is reduction to
Area Scoring and then adjustment of at least the most frequent differences:
http://home.snafu.de/jasiek/diffasts.html
http://home.snafu.de/jasiek/asintro.pdf
--
robert jasiek
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/