Author: Tom Kerrigan
Date: 04:30:39 12/29/01
Go up one level in this thread
On December 29, 2001 at 06:15:47, Steve Maughan wrote: >backpropagation training as it may trigger some thought processes. Another idea >may be to run this for a large number of positions (~10,000) with a shallow >search (4 ply) and collect all of the coefficient deltas - applying the changes >at the end. This may give a more stable situation. I don't think you need a Possibly the most important part of the idea is that the searches are iterative, so this suggestion will not work. Let's say the program is shooting for position X, but _should_ be aiming for position Z. The algorithm makes X less attractive and Z more attractive, but that might make the program try for position Y instead of Z. By making only minor adjustments to the weights and iterating the searches, the program's behavior is disturbed as little as possible while still making it play the correct move. -Tom
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.