Author: Robert Hyatt
Date: 10:48:18 09/06/00
Go up one level in this thread
On September 06, 2000 at 10:49:03, James T. Walker wrote: >On September 06, 2000 at 10:35:38, Robert Hyatt wrote: > >>On September 05, 2000 at 23:27:23, James T. Walker wrote: >> >>>On September 05, 2000 at 16:28:34, William H Rogers wrote: >>> >>>>On September 05, 2000 at 14:37:16, Vladimir Sokolov wrote: >>>> >>>>>It is noticed that many chess programs use automatic learning. >>>>>If an operator lets his computer run at night, in the weekends and vacations, >>>>>then after a few weeks, months the learn files will become stronger. >>>>>Can a weak free program this way become stronger than a raw strong commercial >>>>>program ? >>>>>Can programs after a while use moves, lines and strategies that they did not >>>>>know or were not in their memory or book in the beginning but learned from other >>>>>programs by playing against them ? >>>>>Are these moves automatically imported in their books or must the operator >>>>>import them or run the analysis of the games ? >>>>>Are there programs running on the chess servers that continuously automatically >>>>>play against all challengers and how can players import the learning of those >>>>>programs into their own programs ? >>>> >>>>It my understanding,and I may be wrong, that automatic learning only applies to >>>>the programs opening books, that is their opening books get better with >>>>automatic learning, and not the program. Some programs can share books but not >>>>all. Very few, if any, programs change their logical evaluations based upon a >>>>learning module. If I misunderstood you post, I am sorry, if I did understand >>>>then I hope that the above explains it to you. >>>>Bill >>> >>>I believe some programs also have learning functions which apply to the middle >>>game/endgame. Hiarcs comes to mind. It has a file for this which grows to 64K >>>and then stops growing. I assume it then ejects the oldest positions learned >>>for the newer ones. >>>Jim >> >> >>Position learning really won't help in endgames, unless you have a "magic >>bullet" you can use to reach the same endgame over and over. In chess, this >>would be nearly impossible to do. >> >>It does help when you get taken out of book, at the same point, repeatedly... >>That is the reason I implemented it myself. IE if someone plays 1. a3 and >>2. h3, everybody will be out of book quickly. Now you can re-play the same >>game over and over (on a server) and clean the program's clock by doing so. >>With position learning, it still will vary enough that this won't work very >>well. > > >Thanks for your input Bob. I'm sure you are right about the endgame part. Also >I believe it was Nimzo 99 that has the learning file and not Hiarcs. I believe >Crafty also has something similiar. >Jim Yes. Dave Slate and Tony Scherzer wrote a paper in the JICCA (about 10 years ago IIRC) that described exactly what I implemented in Crafty by following their approach.
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.