Author: Peter Fendrich
Date: 16:40:09 01/02/01
Go up one level in this thread
On January 02, 2001 at 18:52:46, Christian Söderström wrote: >On January 02, 2001 at 17:53:41, Peter Fendrich wrote: > >>On January 02, 2001 at 16:10:53, Christian Söderström wrote: >> >>>On January 02, 2001 at 15:36:45, Jon Dart wrote: >>> >>>>On January 02, 2001 at 05:30:07, Christian Söderström wrote: > ><---> > >>>So isn't it possible to supplement your idea by also >>>learning from the end-result of a game? >>> >>>- Christian >>> >>>>--Jon >> >>That's what I am doing. >>Very shortly: >>I have a book file and a "post" book file. The first one is build from >>pgn-files. The latter one is build by Terra itself when saving the 10 or so >>first moves after leaving the book. The game result is affecting the values for >>all the moves used from the book and Postbook. >>I combine this with a more complex logic for Postbook that backs up part of the >>evaluation a couple of plies upwards in the played line. The back up value is >>decreased for each upward step. > >Sounds interesting! But how do you treat moves in the postbook? >Can they be trusted as much as moves in the regular book? The Book file is used if there are any moves with a value above the "avoid treshold" (see the kasparov example below). The Postbook is only used when Terra is out of book. If I find a "good" move there, it will be played. If not I'll create a new postbook entry with the new computed move. In both cases the played move will be updated dep. on later evaluations and the game result. The value is set so that if the same game result is repeated that will dominate over any backed up evaluation. > >>The last complex part is not as useful as it's beautiful! >>It's far to sofisticated. I have to clear the learning completely from time to >>time anyway. New releases plays different. > >True true. > >>Two parameters to consider is the rating of your opponent and the time settings >>(at least when backing up evaluation). >>For example when Kasparov outplays Mint (if he's lucky of course...) > >Well it's theoretically possible I guess. > >>10 games in >>a row Mint might consider 1.e4 and 1.d4 as seriously bad moves and never try >>them again. It's time to clear the learning again ... :-) > >Yeah you're right :) I have a really sophisticated learning >system worked out, based on an x^2 function that will hopefully >work well (read: at all). I can send a parameter to the >function that tells it how much weight to put into this game, >based on probably time control and opp-rating like you say >but probably also how many games has been played with this >opening prior to this. > >Also, it must be recognized if the opponent attained a winning >advantage but goofed up, that should be treated as a loss. I don't even try to do this. Lower rating and shorter time frames means lower impact on learning and higher possiblity of goofing up. Higher rating and more time is the oher way around. It is kind of covering your case in some way I think. //Peter
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.