Ticket #213 (closed defect: fixed)
Node limit bug and failure to make seki.
| Reported by: | gunnar | Owned by: | gnugo |
|---|---|---|---|
| Priority: | normal | Milestone: | 3.9.1 |
| Component: | source | Version: | |
| Severity: | normal | Keywords: | |
| Cc: | patch: | yes |
Description
From http://lists.gnu.org/archive/html/gnugo-devel/2010-02/msg00013.html
Good morning, Sir
I am a Wei Qi (Go) player and I love gnugo go.
I use gnugo 3.8 to practice dead-live problem as attachment file,
gnugo plays as white.
The command line is: gnugo --mode gtp --capture-all-dead
--chinese-rules --monte-carlo --level 100
1. The final position is wrong, gnugo should pass and the white
can be a both-live, i suppose it a bug.
notes:
2. Though in the docs there are no level 100, if i just set as
level 10, the first round gnugo will resign.
3. if i don't set monte-carlo option, so the first round gnugo
will play the wrong position.
Wish it useful for gnugo to become stronger and I hope gnugo can
play absolutely correct at small board.
BTW, the first of black is not the best choice, I just want to see
if the opponent can't play correct, is it possible gnugo can win. The
best choice for black can kill all white.
If it is really a bug please inform me, I will be very happy. And
if just add some pattern database, please tell me how to repair it,
Thank you very much!
Attachments
-
bug.sgf
(219 bytes) -
added by gunnar 2 years ago.
- Failing position attached to original e-mail.
-
gunnar_9_1.2.diff
(1.8 KB) -
added by gunnar 2 years ago.
- Fix for overflow in node limit computations.
-
testcase_9_1.1.diff
(3.5 KB) -
added by gunnar 2 years ago.
- Two testcases for move 2.
-
gunnar_9_1.3.diff
(526 bytes) -
added by gunnar 2 years ago.
- owl tuning
-
gunnar_9_1.4.diff
(1.2 KB) -
added by gunnar 2 years ago.
- Attempt to revise heuristics for performing semeai analysis for inessential dragons.
-
gunnar_9_1.5.diff
(587 bytes) -
added by gunnar 2 years ago.
- Another attempt to revise inessentiality heuristics.
Regression Results
| Attachment | Rev. | PASS | FAIL | Nodes | Status | |
| gunnar_9_1.2.diff | never tested | |||||
| gunnar_9_1.3.diff | 2440 | 1 | +0.00025% +0.008% -0.00057% | builds with warning(s) | details | |
| gunnar_9_1.4.diff | 2442 | 1 | 3 | +0.03% +0.081% +0.017% | details | |
| gunnar_9_1.5.diff | 2442 | 1 | -0.013% +0.0051% +0.00025% | details | ||
| testcase_9_1.1.diff | never tested |
Change History
comment:1 Changed 2 years ago by gunnar
As discussed at length in http://lists.gnu.org/archive/html/gnugo-devel/2010-02/msg00017.html there's an overflow bug in the computation of node limits causing the problem at the end of the game. At move 2 there's a semeai reading mistake which can be fixed by pattern tuning, but it's not sufficient to obtain the right move.
Changed 2 years ago by gunnar
-
attachment
gunnar_9_1.2.diff
added
Fix for overflow in node limit computations.
comment:2 Changed 2 years ago by gunnar
- patch set
The reason why fixing the semeai reading mistake isn't enough is that A2 is considered inessential and thus ignored for semeai analysis. Some heuristics need revision.
Changed 2 years ago by gunnar
-
attachment
gunnar_9_1.4.diff
added
Attempt to revise heuristics for performing semeai analysis for inessential dragons.
comment:3 Changed 2 years ago by gunnar
- Milestone changed from 3.9.x to 3.9.1
The patch gunnar_9_1.4 has a surprisingly small impact on the regressions
filllib:14 FAIL B1 [PASS] global:42 FAIL A8 [M4] semeai:153 FAIL alive [dead] semeai:155 pass
but all the three failures are very bad so the patch is unacceptable.
Changed 2 years ago by gunnar
-
attachment
gunnar_9_1.5.diff
added
Another attempt to revise inessentiality heuristics.
Failing position attached to original e-mail.