[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] findcap()
From: |
Evan Berggren Daniel |
Subject: |
[gnugo-devel] findcap() |
Date: |
Fri, 22 Nov 2002 14:21:22 -0500 (EST) |
This patch generalizes findcap3() into findcap(), which works on strings
of 4 liberties now. Attack4() now calls findcap() after trying other
moves.
The motivation for the patch was to fix capture:4. Before the patch, the
correct move is not considered. After the patch, it is considered, but
after the sequence b H6, w F6, b F7, w E1, B F2, w H2, the reading is too
deep and the move at J2 is not found. The test passes with 3 calls to
increase_depths() first.
The result is a small delta (2 PASS, 3 FAIL) with a general improvement,
and a decrease in reading nodes in reading.tst from 97435 to 97293, and in
owl.tst from 31160705 to 30671831, and owl nodes in owl.tst from 22453 to
22395.
Thanks
Evan Daniel
- generalize find_cap3() to strings with more than 3 liberies
- attack4() now uses find_cap() to look for attacking moves.A
The delta:
ld_owl:302 PASS properly solved
optics:1201 FAIL improvement
arb:150 FAIL maybe real
golife:4 FAIL improved
lazarus:3 PASS properly solved (?)
optics:1201
This is the same problem attack5 caused. The tactics code now sees an
attack and a defense for the T13 dragon, however the owl code believes it
is dead. I suspect a problem with marginal eye vertices, and plan to
investigate further.
arb:150
Owl no longer finds J5 as an attack move. However, with depths increased
by 1, it continues passing. I have not investigated in detail.
golife:4
I think that gnugo's understanding of the situation has improved; A3 is
now alive, not dead, which is an improvement. However, the correct move
at H5 is only valued at 6 points, which I believe is too low.
- [gnugo-devel] findcap(),
Evan Berggren Daniel <=