On 19x19, Many Faces has books.  A full board opening book made from strong
player games is a hash of all positions (rotation/refection invariant).  It
keeps statistics of player strength and win rate, and is only used to bias
the search, not to choose a move quickly.

That's what I re-invented recently, and I combined it with a joseki book made from strong 
player games as well. Both books just recommend a bunch of moves, the appropriate choice 
is left to the search algorithm. Because I don't return a move instantly without search, 
I call it "active book application".

So far (without any kind of parameter tuning or manual changes to the book) it 
increased Orego's win rate against GNUGo by 4-5 percent, with 10 seconds per 
move. I wonder if the effect would be stronger or weaker for a stronger program 
- it's more difficult to improve a stronger program of course, but it might 
also understand some of the openings better and make better use of them.

A clean, hand-coded book of joseki might be very strong in this framework, but 
I don't have the time.

Thanks everybody for your answers. I might post a short paper about this soon 
if people are interested.

_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to