Author: Tom Kerrigan
Date: 19:00:42 12/29/01
Go up one level in this thread
On December 29, 2001 at 19:03:20, Sune Fischer wrote: >On December 29, 2001 at 17:17:54, Tom Kerrigan wrote: > >>Genetic algorithms are only good if the evaluation (of strength) function is >>fast. In chess, you have absolutely no hope of making this fast. >> >>-Tom > >You have a large search space so you will _always_ need many evaluations. >Genetic algorithms converge relatively fast, and it does a global search unlike >the steepest decent scheme of backpropagation. If you do it right genetic >algorithms can converge faster than backpropagation (well at least for some >problems eg. feed forward networks). > >Anyway, there are lots of ways to fit the parameters once you have something to >fit them by, getting the eval and getting it cheap is the problem. Right, I see it as an insurmountable problem. If there were a quick way to accurately determine the strength of a chess program, all of our programs would be rated 4000 FIDE... -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.