Re: [gnugo-devel] thrashing dragons again

[Gunnar Farnebäck]

> There are 2 changes:
> 1. Dont make it dependent on the size of the move value without
> thrashing dragon heuristic.
> 2. Don't make it dependent on being ahead.
> 
> I suppose you only disagree with 2., and it is certainly debatable.

To really do it right one should probably include current score, size
of moves elsewhere, and how unsettled the board is. For GNU Go I think
it's fine to generate restraining moves anytime we are ahead, except
maybe in ko fights. Even if it's not necessary to win a particular ko
to win the game it would look silly to regard every thrashing move as
a ko threat. Hm, how are we handling that currently?

> My change was motivated by the fact that when a thrashing dragon
> move occurs on kgs, I almost always want gnugo to answer. So my bet
> is that it is a win more often than not.

If GNU Go is too far behind to catch up anyway I suppose it looks
better to lose gracefully than to collapse somewhere. But in that case
we would again need an estimate of the size of moving elsewhere to
determine if we have a chance to catch up by ignoring the thrashing
dragon.

> I don't know. But it would be sufficient if one of the correct moves
> would get a strategic attack reason against C6; since their territorial
> value is much higher than A4, they will then be played for sure.

That's why I suggested a c pattern for H3 but when thinking about it
that would only generate a strategical defense. On the other hand it
wouldn't be completely unreasonable to add bonus for strategical
defenses of surrounding dragons as well.

> (That's why I gave the strategic attacks exactly the same bonus as the
> connection moves. I think this is actually the most successful part of
> my patch.)

I have some recollection that this didn't work that well long ago, but
many things have changed since then.

/Gunnar