P.S. As for correct komi, I'm convinced that 8pt or 9pt favours white.
I'm not yet convinced that 5pt or 6pt (in Japanese rules) favours black.
But I agree that perfect komi is likely to turn out to be 7pt.
Magnus, do you have an opening line for 5.5pt komi that you think is a
certain black win?
No, I would need a lot of time before making any such bold statements.
As was pointed before here on the list it is a statistical argument
that influences my beliefs about 9x9 komi. If Valkyria as Black starts
with E5 on games on LG seems to win comfortably but it is very hard to
say which moves are "winning moves" or "playable". With white Valyria
seems to be pushed hard for many moves until it slowly gains
confidence in winning (it is undefeated for 50 games or so). 9x9 is
sufficiently complex that one can play slightly inferior moves and
still win even against strong humans (a purely subjective feeling).
Zen and Crazystone also beats Valkyria with black on CGOS with 7.5
komi so not even "high quality" book seems to be even close to optimal
play. At least the book is not deep enough to ensure an advantage
against stronger programs. It could be that I put too much effort on
variations that are never played, so the effective book against strong
programs are to small to have an impact. Losses is also due to the
difficulties in the middlegame and endgame that has a huge complexity.
Some positions in the middlegame are straightforward and the
variations are mostly forced, but I have encountered positions that
are so difficult that they seem completely "unforced" and the
branching factor for playable moves are higher than in the opening.
I am rambling here so basically I am clueless!
The 5.5 book for LG is basically just the collection of games played
on LG. Valkyria has played 117 games and only a handful of them was
against players that would have any kind of chance beating it. That is
whatever winning lines I have in the book it may just be an illusion.
I also think self play to get more strong lines for both colors will
also leave a lot of undetected holes in the book so I try to stick to
just updating this book with real games against real opponents.
One interesting idea is to check if E5 is the the only winning move or
if for example C3 would also be winning for black. Even if it is I
think in practice it would lead to 50% winning chance against an equal
strong opponent.
On LG I played such non E5 moves against low rank players to explore
the opening moves but still not risk losing. Against stronger
opponents I let Valkyria play E5 as black.
Best
Magnus
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
