I think in general this kind of shallow search tree could be great idea. Interesting issua is,how much you branch at each stage. So instead looking ahed 16 moves with as 16x1+x1+x1. How about 8x2x1x1x1 or
2x2x2x2?
It is used by backgammon program. But there it makes well sence since their static evaluation is already at the level of international tournament player. In GnuBg you can set how many of best moves are to be taken into condideration at the next ply and also some additional criteria. For example GnuBg thinks taht move is about 5% worse than an another it very unlikely that this judgement would change by deeper search. In backgammon the serach tree is just as bad as it is Go. But luckily the evaluation is quite bit easier
With few plys of seacr I think all neural network BG-programs could have a even match with world champion.
Petri Pitkänen
Heart disease is Britains biggest killer. Join the British Heart Foundations Big Red Fightback: bhf.org.uk/fightback
_______________________________________________ computer-go mailing list computer-go@xxxxxxxxxxxxxxxxx http://www.computer-go.org/mailman/listinfo/computer-go/