On Wed, May 13, 2009 at 2:44 AM, Magnus Persson <magnus.pers...@phmp.se>wrote:
> Some quick comments: > > I did store the search tree with early versions of Valkyria, but then I > gave it up. > > Problems: > > 1) Searching deeper did not seem to overcome inherent fuseki weaknesses > 2) The memory cost became too high Yes, the memory cost is high. When I started talking about this, I knew that it would involve a lot of space. And you have to stop when you run out of disk space and probably even memory space. In point 1, I think it's clear that program play very weak on the first few moves, so even at super high levels it's difficult to overcome this. > > > Advantage: > 1) Playing fast in the opening saves time. This is very good on small > boards. > > Currently, I am thinking of doing something similar to a normal opening > book but with winrate statistics from played games that are updated online. > The idea is to bias search towards known good moves, but avoid: > > 1) Memory costs > 2) The problem of overfitting weak and strong opponents if only the most > played moves are stored and no search is done in the positions found in the > opening book. Sounds like a reasonable approach to me. The principle is to try to benefit somehow from past experiences and your idea attempts to do that. > > > -Magnus > > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ >
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
