[Computer-go] kgs-genmove_cleanup again
pasky at ucw.cz
Sat Mar 27 17:19:38 PDT 2010
I have yet another query about kgs-genmove_cleanup - I think its
behavior now is rather user-unfriendly and also a bit confusing.
The genmove cleanup command is defined "do not pass until there are no
dead groups on the board". So far in Pachi, I have interpreted that as
"do not pass if scoring requires any group to be marked dead" - this is
trivial to implement, just try to check if the board is a win if all
stones are alive. However, a KGS bug(?) creates a problem:
(1) Bot plays white, you play black.
(2) Get into the cleanup phase.
(3) Pass while there are some dead stones of both colors.
(4) Pachi scores the board as a win for white even if _both_
black and white dead stones are considered alive, thus also passes.
(5) KGS issues a final_status_list and Pachi replies with the
list of dead stones. KGS complains a bit with
WARNING: After board cleanup, engine still claimed dead stone at o10
(6) KGS waits for the player to mark the dead groups and press Done.
(7) KGS now considers apparently the _intersection_ of player's
and Pachi's selection as the final status list.
(8) Since the player marked all white dead groups but only some black
ones, the game is incorrectly scored as a win for black.
(Ex.: pachIW vs joewong 2010-03-27 14:16:32)
I admit that (4) is not entirely correct, and it would be easy to
modify Pachi to pass correctly. However, I think KGS definition of
genmove_cleanup is user-unfriendly since it forces any game that got
into the cleanup phase to be played out completely, while my definition
_in addition_ allows the user to "try to start the scoring again" anytime
and AFAICS should not break programs following the original definition.
I think that (7) is the real problem; if KGS would simply enforce the
engine's status choice like anywhere else, I think everything would work
beautifully; am I missing something? Would it be possible to fix the
server or should I adjust Pachi?
[P.S.: Resent with the correct Cc. I shall also mention that Dan Reish
pointed this problem out.]
Petr "Pasky" Baudis
http://pasky.or.cz/ | "Ars longa, vita brevis." -- Hippocrates
More information about the Computer-go