[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] Influence patch
From: |
Gunnar Farneback |
Subject: |
[gnugo-devel] Influence patch |
Date: |
Wed, 14 Nov 2001 20:17:14 +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 patch at
http://www.lysator.liu.se/~gunnar/gnugo/patches/gunnar_1_14.7.gz
This patch makes some badly needed improvements to the influence code
and some tuning.
- non-territory patterns introduced in the influence code
- revised formula for addition of followup_value
- better trace output for followup_value and reverse_followup_value
- tuning
- owl tuning
- eye tuning
- influence tuning
- test case revisions
The main problem with the influence code that this patch solves is
that false eyes counted as territory, e.g. in this position
OOOXX
O..OX
OOOXX
O was thought to have two points of territory and after capture
OOOXX
O.X.X
OOOXX
X was thought to have one point of territory plus the captured stone,
making the capture valued to 4 points instead of 2.
Unfortunately it now thinks that all false eyes are worthless, so it
will undervalue moves like * here (X is assumed to be strongly alive
on both sides)
..O..
XXOXX
..*..
-----
This is not so easily solved though since it requires a detailed
analysis of the positions of eyes. For a good example, compare the
value of * with and without the x in place in this position
XXXOOOO....
X.XOOO.OO..
XX.XXO.O.OO
.XXX.XOOO.O
OOOOX.XO.OO
.O.OOX.XXOO
.OO.O*X...O
..OOO.XxXXO
....OO.O..O
.....OOOOOO
The patch is already in CVS.
/Gunnar
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [gnugo-devel] Influence patch,
Gunnar Farneback <=