[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] reverse followup valuation
From: |
Paul Pogonyshev |
Subject: |
[gnugo-devel] reverse followup valuation |
Date: |
Tue, 12 Aug 2003 22:00:51 +0000 |
User-agent: |
KMail/1.5.9 |
i have recently spotted quite a few positions where reverse
followup is completely inadequate. here are two very common
cases:
XXXX
XO.O
XO.O
XXXX
in this position the moves at '.' are avarded large reverse
followups because they prevent an atari. i can see no way to
prevent such positions from matching in `Sente13' pattern.
i'd propose the following solution: start a pattern database
named say `invalidate.db' which will remove inappropriate
followups/reverse followups/anything else. the reason to
have a separate database is that we need to be sure that
invalidation patterns match after all other patterns. first
pattern in this database would look like
Pattern Invalidate1
O*O
O.O
:+,O
A*B
AaB
; lib(A) == 2 || lib(B) == 2
> remove_reverse_followup(*);
> remove_reverse_followup(a);
second common position is like this:
OXXXXX
.OO*..
------
the move at '*' is avarded a reverse followup because it
"prevents" atari. actually, it only prevents atari from
happening at '*' and moves it one/more vertices to the right.
the proper solution would be to improve defend_against_atari()
in `helpers.c'. i'd suggest something like this:
IF move is adjacent to the string we are "saving" from atari
AND after the move the string still has two liberties
AND opponent has a safe atari on the string after the move
THEN no reverse followup should be added
maybe the first condition is not necessary because of not so
common third position:
O*XX
O.OX
O.OX
OXXX
here move at '*' receives a reverse followup because it prevents
atari on the second white string. but actually, it doesn't -
it only prevents one of two possible safe ataries.
i'd like to hear comments on this. reverse followup valuation is
important, especially in endgame.
Paul
- [gnugo-devel] reverse followup valuation,
Paul Pogonyshev <=