gnugo-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[gnugo-devel] Updated large scale patch


From: Stéphane Nicolet
Subject: [gnugo-devel] Updated large scale patch
Date: Mon, 11 Aug 2003 01:57:31 +0200

Hi all,

Here is a new version of the large scale attack patch.

The regression results are now 11 FAIL and 22 PASS, compared
to 19 FAIL and 13 PASS in the previous version. I have tried
to reduce the problem of the bad aji left by the large scale
captures, following the advices of SP LEE and Arend on the list
(many thanks to both for their comments, btw).


- As (almost) proposed by Arend, the owl reading for tentative
  large captures is now dynamicaly limited to 350 nodes for
  moves at distance <= 1, and 150 nodes for moves at distance
  2 or 3. This should tends to limit the search to moves that
  kill cleanly (ie leave less aji), and also help a little
  bit with the speed since some of the owl searches for failed
  attacks terminate faster (in 150/350 owl nodes instead of 1000).


  Of course, I introduced a new function change_owl_node_limit()
  in owl.c and a new field owl_node_limit in each persistent owl
  cache entry. The result stored in the owl cache is now trusted
  if it is marked certain or if the owl_node_limit used when the
  result was put in cache was greater than the current owl_node_limit.


- correction of a bug in the calculation of the physical
  extention of the dragons, which prevented the patch to apply
  in 1/4 of the cases where it should have.

- evaluation malus for suspected ineffective owl attack of worms
  in estimate_territorial_value() set to 2.0+epsilon instead of
  epsilon.


I think the worse problem of that patch at the moment is the
speed. Arend posted some benchmarks yesterday, but I think
that computer/computer games are best cases for the patch
since they tend to be calmer than human games against gnugo.
When almost of the board is in fire the patch will try quite
a lot of owl-attacks against small dragons.

Ideas on how to further reduce the FAILs or how to speed up
the patch would be most interesting, of course. Or any comment,
for that matter :-)

Stephane.


********

39 unexpected FAIL: Correct 'R7', got 'S4'
    http://www.public32.com/regress/?nngs1.tst:39

2 unexpected FAIL: Correct 'Q18|S16', got 'S15'
    http://www.public32.com/regress/?arion.tst:2

13 unexpected FAIL: Correct 'R13|M8|L9', got 'J3'
    http://www.public32.com/regress/?lazarus.tst:13

89 unexpected FAIL: Correct '(D16|B7|G13)', got 'H11'
    http://www.public32.com/regress/?strategy2.tst:89

163 unexpected FAIL: Correct 'B9', got 'C7'
    http://www.public32.com/regress/?trevor.tst:163

700 unexpected FAIL: Correct 'K17', got 'K16'
    http://www.public32.com/regress/?nngs.tst:700

22 unexpected FAIL: Correct 'F2', got 'H2'
    http://www.public32.com/regress/?global.tst:22

200 unexpected FAIL: Correct 'P6|P7|Q7|S18', got 'S8'
    http://www.public32.com/regress/?strategy4.tst:200

110 unexpected FAIL: Correct 'B14|B13', got 'M16'
    http://www.public32.com/regress/?nngs2.tst:110

490 unexpected FAIL: Correct 'P15', got 'T13'
    http://www.public32.com/regress/?nngs3.tst:490

4 unexpected FAIL: Correct 'F4', got 'K1'
    http://www.public32.com/regress/?13x13b.tst:4


42 unexpected PASS!
    http://www.public32.com/regress/?nngs1.tst:42

6 unexpected PASS!
    http://www.public32.com/regress/?arion.tst:6

80 unexpected PASS!
    http://www.public32.com/regress/?strategy2.tst:80

1107 unexpected PASS!
    http://www.public32.com/regress/?nicklas1.tst:1107

1406 unexpected PASS!
    http://www.public32.com/regress/?nicklas1.tst:1406

290 unexpected PASS!
    http://www.public32.com/regress/?nngs.tst:290

380 unexpected PASS!
    http://www.public32.com/regress/?nngs.tst:380

1160 unexpected PASS!
    http://www.public32.com/regress/?nngs.tst:1160

1300 unexpected PASS!
    http://www.public32.com/regress/?nngs.tst:1300

1750 unexpected PASS!
    http://www.public32.com/regress/?nngs.tst:1750

114 unexpected PASS!
    http://www.public32.com/regress/?strategy3.tst:114

65 unexpected PASS!
    http://www.public32.com/regress/?13x13.tst:65

66 unexpected PASS!
    http://www.public32.com/regress/?13x13.tst:66

202 unexpected PASS!
    http://www.public32.com/regress/?strategy4.tst:202

600 unexpected PASS!
    http://www.public32.com/regress/?nngs2.tst:600

220 unexpected PASS!
    http://www.public32.com/regress/?nngs3.tst:220

470 unexpected PASS!
    http://www.public32.com/regress/?nngs3.tst:470

950 unexpected PASS!
    http://www.public32.com/regress/?nngs3.tst:950

160 unexpected PASS!
    http://www.public32.com/regress/?nngs4.tst:160

260 unexpected PASS!
    http://www.public32.com/regress/?century2002.tst:260

106 unexpected PASS!
    http://www.public32.com/regress/?tactics1.tst:106

40 unexpected PASS!
    http://www.public32.com/regress/?13x13b.tst:40


*****



Attachment: large_scale_5_1.2
Description: application/applefile

Attachment: large_scale_5_1.2
Description: application/text




























... /// ... \\\ ...

reply via email to

[Prev in Thread] Current Thread [Next in Thread]