[gnugo-devel] GNU Go 2.0 and Liberty 1.0

[Aloril]

> This sounds quite interesting. Those programs are real antiques.

They are ;-)
I wanted something that would play at beginner level.

> I played a few games between GNU Go 2.6 and GNU Go 3.7.4 when
> I gave a lecture at PARC before the lecture. I think I started
> with a 4 stone handicap but 3.7.4 was winning every game so I
> tried a 9 stone game and it won that too. I think we believed
> that 2.6 was six stones stronger than 2.0.

I have noticed that if we take 3 bots A,B,C and if:
C+9 stones==B and B+9 stones==A, then its unlikely that C+18 stones==A.
More likely for example something like C+13 stones==A.

This needs more studying but I suspect its general phenomenon.

GNU Go 2.0 wins easily Liberty 1.0 with 9 stones. Not always because
GNU Go 2.0 has premature pass problem sometimes.

However at KGS GnuGo2 is rated at 18k and LibertyBot is rated at 24k.
They have little idle time playing almost constantly:
http://kgs.kiseido.com/en_US/gameArchives.jsp?user=GnuGo2
http://kgs.kiseido.com/en_US/gameArchives.jsp?user=LibertyBot

This rating is quite surprise to me.
I'm rated at 13k at KGS and 13 stones or more seems to be fair handicap
between me and GNU Go 2.0.
Also it seems that 25 stones or more is fair handicap between me and 
Liberty 1.0.

I think that their strength cutting/basic tactical reading is enough
to offset their weakness: l&d, weak strategy, etc..

If opponents knew about their weakness, they would be rated much lower.

> I don't think we object but Nick Wedd might object.

I'm entering then into Open division, it has less restrictions than Formal 
division.
I have talked earlier with him and I think he was OK with it.

-- 
Aloril <address@hidden>