[gnugo-devel] Opponent not passing

[Daniel Bump]

The patch trevor_1_24.3 contained something called
opponent_not_passing with the remark:

/* If we're way ahead, and opponent has not been passing when we
 * think there are not even dame left, let's start playing aftermath
 * moves, too.  This funtion returns 1 if we should play an aftermath
 * or capturing move.
 * See, disaster in games/nngs/gnugo-3.1.22-heeroy-200201252006.sgf
 * See also TurboGo in the 2001 European Congress (Dublin), where GNU
 *   Go took some really unnecessary risks in passing over & over.
 *   http://www.britgo.org/results/computer/egc01/tg-gg.sgf
 */

I don't think calling the aftermath code is the right
approach here.

BTW in the Turbogo game GNU Go's passes don't look
egregious. However endgame tenukis can be real problem.

This was the reason we implemented the owl threats code,
which generated defensive moves even if the opponent was
only *threatening* to do something. Note that owl
threats are taken seriously, tactical threats are
not. This is because the tactical reading code is
trusted but the owl code is less so.

Many Faces has a policy that if the opponent adds
stones to a dead dragon then MF must generate a
move in that area. I think that's going too far
but we do try to do something like that with the
OWL_PREVENT_THREAT move reason. See the comment
around line 1450 in value_moves.c.

This is one of the few places in the code where
the location of the opponen't last move is taken
into account. Other owl threat move reasons generate
moves with low value that attack a group that can
threaten to live, or that threaten to revive a
dead group. The idea is to generate such moves with
a low value so that they won't be played unless
we're down to the penny ante stuff.

I think that what should really be done is getting the
owl threat code turned back on near the end of the game.

Arend has been hacking the owl threats and might have
an opinion on how reliable it currently is in preventing
endgame tenukis through overconfidence. We've seen
many cases where it can contribute a bad or pointless
move, but if we're ahead, safety is the best policy.

The owl threats were turned off recently because the
engine was becoming too slow, but near the end of
the game the slowness of the engine is no longer
a problem and if we could figure out a suitable
criterion for turning it back on that might be the
best approach.

Dan