Author: Bruce Cleaver
Date: 01:15:37 08/21/98
Go up one level in this thread
On August 20, 1998 at 12:23:33, Torsten Schoop wrote: >Hi all, > >does anyone know why NN backgammon programs play very well (TD-Gammon, >JellyFish, Snowie)but the NN chess playing programs do not? > >Ciao >Torsten One reason that minimax is NOT successful in Backgammon is that the branching factor is enormous (about 400 per ply, with dice throws) so a normal alpha-beta search may only be able to get 2 or 3 ply on even a fast machine. So, other methods had to be found. TD-Gammon went through several revisions, first with a raw board input, then with hand coded primitives which substantially improved its play. The equivalent in chess would be normal eval stuff like passed pawns, hanging pieces, king safety, good and bad squares, material advantage. TD-Gammon also played an enormous number of practice games to tune its NN (several hundred thousand, or a few million, I cannot remember). I bet if this sort of effort were applied to chess, it could be successful. Bruce
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.