Remove owl vital attack pattern VA44

[gunnar]

Pattern VA44 is terribly general and is likely to prolong owl reading unnecessarily by introducing doubt about good eyespaces really being worth two eyes. This patch simply removes the pattern.

[gunnar]

Remove VA44.

[gunnar]

Mixed regression results but good effects on the node counts.

trevora:290     FAIL C8 [!C8]
global:35       FAIL B13 [M12]
vie:36          FAIL 0 [1 M12]
STS-RV_e:102    PASS 1 0 PASS [1 0 PASS]
owl1:319        PASS 1 S6 [1 S6]
nngs3:140       FAIL G9 [D9]
nngs4:230       PASS Q18 [Q18]
nngs4:900       PASS Q9 [!Q18]
strategy5:292   FAIL S4 [F16|F17]
arend2:210      FAIL N3 [K2]
owl1:397        pass
owl1:354        pass
6 PASS (4 PASS, 2 pass)
6 FAIL
Total nodes: 1680248828 3213584 12373730 (-1.2% -4% -0.24%)

trevora:290 is mostly noise, the current move at E6 isn't all that convincing either.

global:35 and vie:36 are effectively the same and very serious.

[gunnar]

Try to keep VA44 but with a harder constraint.

[gunnar]

Fixed previous patch to actually do what was intended.

[gunnar]

arend2:210 is actually a slight improvement because the only change in owl results is that owl_does_defend N3 M2 returns WIN now, although the reading is pretty much bogus both before and after the patch. The move valuation misses the point completely in both cases since the owl reading of H3 goes wrong.

[gunnar]

strategy5:292 is also regression noise. The change is that R3 is thought to be owl attackable with ko at S4 (rather than safe). This may or may not be correct but at least it gets farther in the reading before running out of nodes, not playing lots of pointless moves caused by VA44. The relative scoring of the upper left and lower right is almost certainly wrong but that's a different problem.

[gunnar]

ninestones:150 is similar to strategy5:292. An new owl attack on G15 is found with the patch gunnar_7_13.6c. The attack may or may not be correct but at least it doesn't run out of nodes by playing nonsense moves suggested by VA44.

[gunnar]

owl1:319 is a slight improvement.

STS-RV_e:102 is a very minor improvement.

nngs4:230 avoids running out of owl nodes, thereby getting the right result.

[gunnar]

nngs4:900 is also a minor improvement.

[gunnar]

owl1:354 also avoids running out of owl nodes.

[gunnar]

Summary of regressions for gunnar_7_13.6c:

trevora:290     FAIL C8 [!C8]
STS-RV_e:102    PASS 1 0 PASS [1 0 PASS]
owl1:319        PASS 1 S6 [1 S6]
nngs4:230       PASS Q18 [Q18]
nngs4:900       PASS Q9 [!Q18]
strategy5:292   FAIL S4 [F16|F17]
ninestones:150  FAIL G14 [C5|D5]
arend2:210      FAIL N3 [K2]
owl1:354        pass
5 PASS (4 PASS, 1 pass)
4 FAIL
Total nodes: 1683336446 3221992 12382463 (-1.1% -3.7% -0.17%)

Some minor improvements, no significant failures, and a substantial reduction in owl nodes.