Re: [gnugo-devel] Patch for break_chain3_moves

[SP lee]

>
> Yes, it's a real problem, slowing down the engine substantially for an
> unclear gain.
>
> /Gunnar
>

So it seems better not touch the various depth parameters. I have modified
my patch to be effective only on not-deep stack levels and now I got the
following regression changes.

./regress.sh . reading.tst
64 unexpected FAIL: Correct '3 N1', got '1 R3'
194 unexpected PASS!
./regress.sh . strategy.tst
13 unexpected PASS!
./regress.sh . arb.tst
203 unexpected FAIL: Correct 'T7', got 'N3'
./regress.sh . lazarus.tst
5 unexpected PASS!
./regress.sh . nngs.tst
370 unexpected PASS! [H17]
371 unexpected PASS!
./regress.sh . 13x13.tst
39 unexpected FAIL: Correct 'H4|J4', got 'J5'
42 unexpected PASS! [N3]
./regress.sh . owl1.tst
263 unexpected PASS!
./regress.sh . century2002.tst
10 unexpected FAIL: Correct 'R15', got 'S17'
60 unexpected FAIL: Correct 'Q14', got 'F10'
./regress.sh . ninestones.tst
40 unexpected PASS! [K19]

I'm a little surprised that blunder:11 is failed on the current cvs version.
The function fast_defense may need to be checked, but owl couldn't find the
right answer either.

For reading:64, it is kind of wasting of ko threats playing on R3, because
the opponent has to answer anyway. It is failed due to the depth threshold
for break_chain2_moves. We could expect that in an actual game, after all
these wasting moves, gg will finally play on N1.

arb:203: Failed probably due to the ko depth threshold.

13x13:39: not really a fail. J5 is acceptable.

century2002:10: not really a fail. S17 makes gg's group thicker in sente,
although the opponent can escape part of the stones. After S17, I played
with gg in the following sequence: B:S17 W:S18 B:R18 W:T17 B:Q18 W:S14 B:R15
W:T15 B:H15. So gg got important sente and played on H15 to strengthen its
weak group.

century2002:60: gg found F10 after added this patch for attacking the left
weak group. To me F10 is bigger indeed. At the same time gg still recognized
Q14 as the defend move for the upper right corner.

To summarize, there is no significant fail due to this patch. Some of the
passes, such as nngs:370, 371 and owl1:263 have corrected really bad
mistakes of gg.


SP Lee

readingpatch.dat
Description: Binary data