[gnugo-devel] inessential problems

[Arend Bayer]

I replayed a few nngs games of 3.1.31/32 with GNU GO 3.1.29 to see whether
there has been a noticeable regression. I did not get any deviation for the
3.1.31 games, but there were several for 3.1.32.

One of them was quite disastrous in the game
gnugo-3.1.32-riceman-200204151811.sgf.
At move 176, 3.1.29 would have played at F10, saving the inessential
stones F9/G9. This is mostly due to some completely illogical territorial
effects around the "saved" inessential stones; but the effect will always
be the same: Give some territorial valuation to saving inessential stones
that are next to an opponent dead dragon.

Not doing this, together with two very bad subsequent owl blunders caused
this dragon (together with another) to revive. This effectively turned a
16 pt loss into a 107pt loss.

Of course, the correct way to fix this is to eliminate the owl failures, the
next best way would a higher strategical value to saving such "inessential"
stones. But the only immediate fix would be to keep the territorial bonus
for saving such inessential stones. This will sometimes eliminate aji from
dead opponent dragons, at the cost of sometimes making -1 point gote moves
in the endgame.

I don't think I am really proposing to make this change backwards before
3.2. Still, I wouldn't rule this out to be a source of effective decrease
in playing strength (which would be more or less impossible to detect in
GNU Go vs GNU Go matches).

Arend