[Pachi] Scoring

Petr Baudis pasky at ucw.cz
Fri Aug 26 11:26:36 CEST 2011


On Thu, Aug 25, 2011 at 10:45:17AM +0100, Gareth Davies wrote:
> I've been using pachi for a short time now.  I ran into a problem yesterday where pachi (with the uct engine) passed very close to the end of a won game only to lose by 18.5 points due to a score counting technicality.  The root of the problem should be obvious from the sgf.

  It is actually not obvious to me. What is the issue?

> This lead me to start digging through the source code and I started to get lost in data and confused by dead groups.  In particular I cannot make sense of the gtp command final_score when run on the result of score.gtp.  Is there some difference between a board state that pachi wants to pass on and a board state that's fit for final_score?

  Here, final_score reports W+10, so it's fine that Pachi passed...?

  Both Pachi passing (unless pass_all_alive option is enabled) and
final_score is based on identification of dead groups. You can check
that using GTP command

	final_status_list dead

				Petr "Pasky" Baudis
UNIX is user friendly, it's just picky about who its friends are.

More information about the Pachi mailing list