Re: [gnugo-devel] Territorial value exaggerated

[Stéphane Nicolet]

SP Lee wrote :

> The first impression is that many moves are not considered by gnugo.

That's probably true in most cases. Note however that --replay only 
report
positive scores, so that it may happen that the move you played was
actually considered by gnugo, but valued negatively : it's quite 
possible
in particular if you played a move considered an invasion by gnugo.


> Then I think gnugo put high value on joseki's and may forget locally
> urgent moves.

I had this impression too, but of course it's a complicated matter how
to balance joseki. One possible solution would be to lower the effect of
minimum_value and maximum_value (as I have done in the cosmic patch
by using weights for them) and see what happens.


> From move 21 to 33 it insists to play on B6, which claims a
> territorial value of 29.27. This really makes me puzzled.
> Even after B6, there are only 2 white stones locally, how
> could they make 29.27 points? I have also seen similar cases
> in other games played by gnugo, where territorial values were
> exaggerated at beginning stage of the game. Can anyone give me
> a clue about this?

This one is easy : the C6 stone is considered alive (or critical)
between moves 13-33. So it diffuses influence, thus negating for
gnugo much of the influence black has on the left edge. That edge
is then neutral for the evaluation function.
So gnugo insist of killing it for good, thus claiming the left edge
as territory (or possibly potential territory, I haven't check if
whose_territory() or whose_moyo() applay in your specific case),
which explains the change in territory value of 29.27 for B6.

Your reasoning "even after another move there, there would be only
two white stones locally, so Black could attack attack the weak group
and consolidate the territory in the upper west edge", is just too
elaborate for gnugo at the moment : it thinks that C6 ruins all
Black territory up to at least C14, probably.


Stephane