gnugo-devel
[Top][All Lists]
Advanced

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

[gnugo-devel] tuning patch


From: Gunnar Farneback
Subject: [gnugo-devel] tuning patch
Date: Sat, 10 Nov 2001 15:12:53 +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)

I've put up a (mostly) tuning patch at
http://www.lysator.liu.se/~gunnar/gnugo/patches/gunnar_1_14.4.gz

- new function is_edge_vertex() in board.c
- guess_eye_space() revised
- compute_eyes_pessimistic revised
- do_owl_defend() revised
- move stupid vital marginal eye points from first line to second line
  in owl_determine_life()
- tuning
- eye tuning
- owl tuning
- bugfix in ld_owl.tst

The main point of this patch is to try to speed up the owl reading
somewhat. This is accomplished partly by being more optimistic in
guess_eye_space(), allowing ending the search quicker, and partly by
pattern tuning. The most important pattern revisions in this respect
is the removal of patterns A413 and D1318 and making the constraints
harder in A408, A411, and A414. This avoids certain nonsensical
hunting sequences in the middle of the open board.

Another change is that the owl code now avoids certain stupid moves on
the first line, e.g. at a in this position:

----+
O...|
OX..|
OX..|
O.X.|
.O.a|
....|

Such moves are instead forced up to the second line. The reason why it
wanted to play ay a is that a marginal eye space happened to end up
there in the eyespace analysis.

Additionally there's a lot more tuning, mostly trying to solve
problems introduced by the owl performance tuning. I believe the
regressions have a net result close to zero due to this patch, but the
speed should be improved.

I've added it to CVS.

/Gunnar



reply via email to

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