Re: [gnugo-devel] Patch for break_chain3_moves

[Gunnar Farneback]

SP Lee wrote:
> I have modified the patch to generate a little less reading nodes and
> also changed the BACKFILL2_DEPTH from 5 to 9, which allows

Such changes must be done with great care since they tend to be quite
expensive.

> break_chain2_moves to be called from attack3 in a rather deep level.
> Now I got reading:64, trevorc:1490, nngs3:710,1170 passed. When I
> increased the ko_depth from 8 to 12 I got nngs4:470 passed. Since
> 13x13:39 is not really a fail, there are totally 9 passes and 3 fails.
> I'll run a full regression test later.

Here are the regression results compared to current CVS:

reading:134     PASS 1 J2 [1 J2]
reading:138     PASS 2 F9 [2 F9]
reading:154     PASS 2 G4 [2 G4]
reading:173     FAIL 0 [1 (A2|C7|B1)]
reading:194     PASS 0 [0]
owl:29          FAIL 1 E9 [1 C7]
owl:31          FAIL 1 E9 [1 C7]
owl:109         PASS 1 C4 [1 (B5|C4)]
ld_owl:193      FAIL critical T4 Q1 [critical (S4|T4|T5) S4]
connection:87   PASS 1 S10 [1 S10]
connection:94   FAIL 0 [1 D9]
trevora:370     FAIL B3 [F6]
trevora:390     FAIL B4 [A2]
trevora:530     PASS H4 [H4]
trevora:531     PASS 1 [1]
strategy:13     PASS N18 [N18]
arb:203         FAIL N3 [T7]
golife:5        FAIL F2 [H5]
viking:2        PASS E2 [E2]
lazarus:5       PASS R13 [R13]
lazarus:13      PASS R13 [R13|M8|L9]
trevorb:120     FAIL E2 [N2]
trevorb:140     PASS K2 [!H4]
trevor:320      FAIL G1 [J2|H3]
tactics:3       FAIL N17 [M18]
nngs:120        PASS M2 [!R5|Q4]
nngs:330        FAIL D15 [B15|B16]
nngs:370        PASS P3 [!Q1]
nngs:371        PASS 0 [0]
nngs:480        PASS G14 [G14]
nngs:1060       PASS F3 [H3|F3]
trevorc:910     FAIL E1 [H13]
trevorc:970     FAIL C13 [J4]
trevorc:1540    FAIL L10 [M10]
trevorc:1580    FAIL B8 [C8]
connect:78      PASS 0 [0]
global:4        FAIL F5 [Q6]
global:5        FAIL F5 [O4]
vie:4           PASS 1 G2 [1 G2]
13x13:42        PASS N3 [N3|N2|L4|L3|L2|L1|M1|N1|M4]
semeai:34       FAIL 1 1 S19 [1 1 S18]
strategy4:151   PASS T5 [T5]
strategy4:155   FAIL C16 [D18]
strategy4:179   PASS E2 [P3|O2|P2|O4|E2]
strategy4:188   PASS B6 [B6|C4]
strategy4:190   FAIL E14 [D13]
owl1:263        PASS 1 R11 [1 R11]
owl1:266        FAIL 1 C18 [1 B17]
owl1:324        PASS 0 [0]
nngs2:150       FAIL J18 [M3|L3]
nngs2:600       PASS B5 [B5]
nngs3:470       PASS G15 [!J17]
nngs3:740       FAIL F2 [!F2]
nngs4:200       FAIL T2 [T6]
nngs4:210       FAIL S5 [!S5]
nngs4:410       FAIL PASS [E1]
nngs4:620       PASS G13 [G13|G14]
nngs4:790       PASS E15 [E15]
nngs4:1040      fail
ninestones:40   PASS K19 [!D2]
ninestones:550  FAIL L9 [G5|J4]
ninestones:570  PASS J15 [A9|B10|D11|F11|J15|A16]
arend2:70       PASS O9 [N7|O9]
nando:6         FAIL 1 [0]
thrash:6        FAIL Q13 [T4|M5|O4|O2|P1]
9x9:40          PASS D4 [D4]
9x9:120         PASS F1 [F1]
9x9:160         PASS E7 [F5|F7|E7]
9x9:250         PASS C1 [A2|C1|C2]
9x9:350         FAIL 1 B4 [0]
cgf2004:50      FAIL N5 [G19]
37 PASS
34 FAIL (33 FAIL, 1 fail)
Total nodes: 2005029175 2870403 11102107 (+34% +2.1% +1.1%)

> I'm not sure whether the increased node count is now a real problem.
> Maybe we could add another level above 10 for experiments.

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

/Gunnar