Actually my text was a little misleading so I updated it too.
Before the iterative deepening method. I had two copies of Valkyria play each other from the position perhaps playing 100 games a day. And then I would use Drago to inspect the winning perentage of the moves played. The two programs would also use a hash table that is updated each game and thus play better and better in theory thanks to learning.
Still I guess this method may be ineffective because so many moves are played in positions very far from root and then not so useful.
On the other hand just as palyouts plays to the end this method can overcome blind spots in evaluation since it play beyound the search horizon.
This version did not use garbage collection, and is also a nice method for book development and overcoming memory limitation. Using just standard tools as twogtp and Drago.
Except for the hash table part it should work with any program that is not deterministic.
/Magnus Quoting Ingo Althöfer <[email protected]>:
Question to the operator of "Fueogo9 bot" on LittleGolem: What are hardware and thinking times? (Best would be to mention it in the account - like done with "Valkyria9 bot")
_______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
