Re: [gnugo-devel] score.c

[Gunnar Farnebäck]

Arend wrote:
> I suggest we stop using Bouzy's score algorithm in score.c. I have run it 
> in comparison with an influence-based score computations while running
> my robot on KGS for a while; while the two are
> often pretty close, the algorithm in score.c is sometimes completely
> off-base in the fuseki, especially in handicap games. In other situations,
> it fails to take into account a good estimate of potential territory.

Fine with me. The only advantage the Bouzy algorithm has over an
influence function based one is speed but in practice it tends to be
insignificant compared to other computations, e.g. the owl reading.

> The score estimate is now done automatically in examine_position(), and
> it sets the new global variables black_score and white_score. I don't like
> global variables either, but I think that's what they are, similar to
> all the dragon_data or initial_black_influence etc.

Agreed. The score estimate should be available in the same way as e.g.
the dragon data.

> --- engine/Makefile.am        15 May 2004 18:13:41 -0000      1.22
> +++ engine/Makefile.am        5 Jan 2005 23:13:26 -0000
> @@ -43,7 +43,6 @@ libengine_a_SOURCES = \
>        printutils.c \
>        readconnect.c \
>        reading.c \
> -      score.c \

Actually I'd like to keep the score.c file around. I have some
experimental code to very quickly guess the score of a finished game
that would make more sense in score.c than anywhere else. Also the
finish_and_score_game() function in play_gtp.c would fit better in
score.c.

/Gunnar