Re: [gnugo-devel] strategic effect

[Gunnar Farneback]

Arend wrote:
> Oops yes indeed. Yet, even with my patch, N11 is still evaluated
> incorrectly; it seems to me that capturing O10 doesn't get the bonus of
> rescueing O9.

That's correct. The analysis of simultaneous threats isn't good enough
to do this yet.

> I assume this belongs to the situation where only a global evaluation of
> a reading result would help. I really do not see a heuristic that would
> help here.

That might be so, but maybe something can be done based on the idea
that it's (often) a bad idea to run away with a dragon when we have
other weak dragons on the board.

> I am sure that GNU Go will improve a lot once readconnect works reliably.
> OTOH I wonder a bit whether it is really useful to have the reading
> analysis split up into one more module. How do you imagine the interaction
> between the owl and the readconnect codes? After all, it is rather the
> rule than an exception these two issues are interdependent. (Make an eye
> by first threatening to cut s.th. off, connect by threatening to capture
> cutting stones, make live with invasion stones by various attempts to
> connect to the outside, reduce eye space while threatening to cut, etc. etc.)

The connection analysis done by readconnect is of a tactical nature.
Just like the tactical reading tries to capture stones by reducing
their liberties to zero, the purpose of readconnect is to try to
connect strings by playing stones to make them actually belong to the
same string. Both of these will be kind of stupid and shortsighted
(tactical reading gives up if the string gets five liberties,
readconnect gives up if it can't find a move getting closer to its
goal).

The strategic understanding of life and death is up to the owl reading
and the semeai reading, which use tactical reading and (in the future)
connection reading as tools to decide which moves are worthwhile to
try.

/Gunnar