[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] invasion question
From: |
nando |
Subject: |
Re: [gnugo-devel] invasion question |
Date: |
Fri, 1 Dec 2006 17:05:19 +0100 |
On 12/1/06, alain Baeckeroot <address@hidden> wrote:
Hi
It seems that GNU Go very often miss "classical" invasions
below 4 th line too streched B stones (like in H9 game)
A quick test seems to show that there is no pattern for such invasion
(...)
Unfortunately, GNU Go isn't fantastically strong at life & death, not
to mention sabaki. So it's probably on purpose that such patterns are
missing.
(...) And i saw a rather big malus for invasion moves (near 10 points)
Correct. See strategic_penalty() in value_moves.c
Is it done on purpose ? (to prevent hasty invasion which fireback)
Exactly.
Is it worth digging, and adding test cases (i found less than 10 invade test)
Possibly. IMHO a serious approach of this issue would have to start
with the creation of a new primitive, something like owl_invasion (and
possibly owl_threaten_invasion). That would allow to more solidly test
pattern-generated potential invasion moves. Then we could use the
result to value the moves more properly (the penalty should still
apply I think). But I doubt a bit that the engine is up to such a task
for the moment.
Loosely related: GNU Go also almost totally lacks reduction moves generators.
-- nando