Re: [gnugo-devel] A terrible game (Patch for Owl tuning)

In the terrible game posted by Dan, two white groups were dead. The attached patch can
save one of the group by tuning owl attack/defense. The other group
involved connection problem and I can't find any solution yet. After
applied the patch, gnugo knows to play S2 at move 63, which is after the
escape route is blocked.

I have modified D209b and added A239 and A240. There are many regression
changes, 6 fails and 11 passes.

strategy.tst 26 FAIL: Correct 'D3' got G3
viking.tst 2 FAIl: Correct 'E2', got 'C4'
lazarus.tst 16 FAIl: Correct 'L16', got 'L17'
strategy4.tst 200 FAIL: Correct 'P6|P7|Q7|S18', got 'S8'
nngs3.tst 280 FAIL: Correct '!D15', got 'D15'
century2002.tst 110 FAIL: Correct 'C8', got 'M2'

strategy.tst 34 PASS: E17
trevor.tst 1060 PASS: 1
trevorc.tst 240 PASS: B2
trevorc.tst 1440 PASS: E12
global.tst 41 PASS: B4
vie.tst 41 PASS: 1 B4 (same case as global:41)
strategy4.tst 192 PASS: A15
owl1.tst 293 PASS: 0
handtalk.tst 10 PASS: E9
nngs2.tst 140 PASS: P5
century2002.tst 260 PASS: Q12

One of the fails I don't understand, i.e. nngs3:280. All I have changed
is owl_attack patterns and owl_defend pattern, why is the result of
influence_delta_territory changed so much?
before : D15: 4.81 - change in territory
after: D15: 3.81 - change in territory
Then D15 is selected because other moves have smaller values in territory as well.

The difference of the moves in other fails and the correct moves seems to be subtle.

I'm checking them now.

SP Lee

From:	SP LEE
Subject:	Re: [gnugo-devel] A terrible game (Patch for Owl tuning)
Date:	Fri, 18 Jul 2003 19:27:08 -0700