Author: José de Jesús García Ruvalcaba
Date: 09:48:54 05/06/99
Go up one level in this thread
On May 06, 1999 at 11:34:51, José Carlos wrote: > I'm finishing writing the first beta version of my chess program (Averno). It >is quite simple at this moment, but I always try to be original, and I have an >idea about the eval function: > Instead of "inventing" rules for evaluating positions from my experience in >chess (or others experience), could it be possible to make the program invent by >itself those rules? I mean, if I was able to write some code for, starting with >a position, an evaluation (taken from grandmasters games) and a best move, the >program could "inference" (sorry, I dont know this word in english correctly, >but I hope you understand) some rules that he, later, use in his games. > How? Well, there are many interesting AI methods to do this. I'm particularly >interested in Neural Networks, but some others may be suggested. > > Have any of you thought of this any time? Any ideas? Any real cases? > > Regards. > > José C. Deep Blue's evaluation was tuned automatically, the weigths of the terms of the evaluation function were solved by a least squares method, trying to reproduce the moves of the winning side of a few thousands grandmaster games. What I do not like about Deep Blue's method is that winning a game against a grandmaster is not a guarantee of playing only good moves in that game. I would like to see a program that actually invents the terms of the evaluation function, then checks if they are actually relevant, and finally weigths them; and in a more rational way than the Deep Blue method. José J.
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.