[Top][All Lists]

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

[gnugo-devel] tuning game

From: Gunnar Farneback
Subject: [gnugo-devel] tuning game
Date: Wed, 21 Nov 2001 18:00:01 +0100
User-agent: EMH/1.14.1 SEMI/1.14.3 (Ushinoya) FLIM/1.14.2 (Yagi-Nishiguchi) APEL/10.3 Emacs/20.7 (sparc-sun-solaris2.7) (with unibyte mode)

GNU Go played an even game against Viking4 on NNGS today, and lost!
This is simply terrible. The main problem was that Viking was allowed
to build a huge moyo while GNU Go picked (too small) territory along
the edges. This is a known weakness which we really have to address
sometime soon. A good start would be to identify test cases from the
game. Below are comments to some of the bad moves. I've left the
fuseki and early middle game for stronger players to criticize. Notice
that we had a chance to win the game close to the end when a
combination attack became possible. Unfortunately we need a better
combination attack module to find it.

The game record is appended at the end and has also been checked in as
regression/games/viking2.sgf. Test cases go into the viking.tst file.

Move 46:
C10 is clearly overvalued. A move to reduce black's moyo seems to be
called for.

Move 52:
Something must be bigger than J18.

Move 70:
E10 looks too small.

Move 88:
The semeai module fools up completely, but the real problem seems to
be that B1 is not classified as inessential.

Move 96:
H13 is clearly bigger than G13.

Move 140:
The combination attack module must be able to find that M14 is
effective. It shouldn't just be played as an empty threat.

Move 146:
Still a chance to win by playing N13 or Q12. This is the
responsibility of the combination_attack module.


PW[GnuGo (GNU Go 3.1.14)]WR[14k*]
C[final_score: B+5.5
F17 removed
F15 removed
C14 removed
M14 removed
E12 removed
C9 removed
R8 removed
K6 removed
B3 removed
H3 removed]

reply via email to

[Prev in Thread] Current Thread [Next in Thread]