[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] failed 3,3 invasion defense
From: |
bump |
Subject: |
Re: [gnugo-devel] failed 3,3 invasion defense |
Date: |
Mon, 28 Oct 2002 08:43:19 -0800 |
> I watched in horror today as gnugo failed to defend against a very early
> and slightly non-joseki 3,3 invasion. I believe this patch to hoshi.sgf
> is the easiest way to fix it.
>
> There may be something better to do about it, but this seemed simplest,
> and it clearly needs to not happen again.
This game features mistakes in 3 corners and might be a
good one to look at.
Evan's patch adds a pattern in hoshi.db:
#maybe possible, but gnugo screws it up with the following sequence
--------+
........|
...*....|
...OXXX.|
....OOX.|
......O.|
........|
........|
........|
:8,sF
>antisuji(*);
I don't like this solution.
GNU plays this move at move 34 because it thinks E2 kills
the B stones in the corner. A better way to fix this would
be to correct this mistake (if it is one) and the mistake
at move 36 by owl tuning.
B cuts on the next move at E4. If W replies at C5 B is in
trouble. B must play at D2 and the corner is still not
alive, though it may be bad to try immediately to kill.
Given that he can win a ladder B can probably get a good
result in this corner after W:E2. Given the W stone at C9
I think B just barely manages making use of a favorable
ladder. So probably E2 does not work but it is close.
I think the real tuning point in this corner is at move 36.
Run gnugo -l gnugo-3.3.10-niki-200210281349.sgf --quiet
-L36 --decide-dragon B4 -o vars.sgf.
After pruning the move tree here's one relevant variation. GNU
thinks B is dead at the end here. Not good ...
(;GM[1]FF[3]
SZ[19]KM[0];
C[owl_attack B4: WIN C1 (attack effective)
(194 variations)
owl_defend B4: WIN C5 (defense effective)
(533 variations)]
AW[cb][eb][pb][dc][kc][nc][dh][ck][nm][rm][bo][oo][qo][cp][dp][eq][er]
AB[cc][ec][fc][qc][cd][pd][de][qf][pj][pm][po][bp][ep][bq][cq][dq][nq][pq]
;W[cs]
C[A403 at B4 (variation 1, hash ca1ca164)
owl_defend B4: 0 (defense failed - genus 0)
(193 variations)
(next variation: 194)]
;B[co]
C[D1112 at B4 (variation 2, hash a52ba8bf)
owl_attack B4: WIN D5 (attack effective)
(100 variations)
(next variation: 102)]
;W[do]
C[save lunch at B4 (variation 86, hash 7f3ee42a)
owl_defend B4: 0 (defense failed - genus 0)
(16 variations)
(next variation: 102)]
;B[cn]
C[D1113 at B4 (variation 87, hash 10ad7ba1)
owl_attack B4: WIN B1 (attack effective)
(2 variations)
(next variation: 89)]
;W[bs]
C[A208a at B4 (variation 88, hash 6ef1137e)
owl_defend B4: 0 (defense failed - genus 0)
(1 variations)
(next variation: 89)]
)
Dan