[gnugo-devel] attack_either patches analysis

[Evan Berggren Daniel]

I finally got around to analyzing the delta on the attack_either patches.

13x13:42     Properly solved
13x13:40     Accidental
trevora:220  unsure, maybe accidental
nngs3:670    Properly solved



13x13:42

The following gtp should make it clear what's going on (fails without the
patch, passes with)

loadsgf games/mertin13x13/katsunari-gnugo2.W+4.sgf 126
black L4
defend_both N7 K4
#? [0]



trevora:220

It's not entirely clear to me what's going on here.  Both before and
after, B6 is seen as a blunder because of the combination attack at C4
that appears.  Move valuation subtracts 4 points as a result.  Before the
patch, this leaves B6 as still the highest valued move (changed from 29.23
tp 25.23).

However, after the patch, a large number of territorial values are lower,
and so B6 starts with a lower value.  When the attack is found, it becomes
one of the lower ranked moves, and another move is generated (changed from
24.74 to 20.74, which is lower than many moves, including C4 which is
eventually chosen).

The territorial value change as large as 4.5 pts leads me to believe that
this not entirely accidental, and is an actual improvement, but I do not
believe the test is properly solved yet.



13x13:40 no longer PASSes, and nngs3:670 passes because of a different
patch.

13x13:40

appears to be small changes in territory values -- move value of K13 is
changed by about .16, so this appears accidental.
(This as compared to 3.3.9, not current CVS)

nngs3:670

here, the reason appears to be that an owl_attack and owl_defend pair is
found for the dragon at D13 that wasn't before.  (D12 is both the attack
and defense point).

As a result, the move value for D12 increases by about 9 points, mostly
due to territorial effects.

loadsgf games/nngs/Lazarus-gnugo-3.1.34-200204280120.sgf 32
owl_attack D13
#? [1 D12]

is the relevant GTP.  I have not investigated in detail how the attack is
found.



Thanks

Evan Daniel