Author: pete
Date: 14:07:07 04/23/00
Go up one level in this thread
I think maybe we see something different here : 0.) Tony Hedlund has previously posted the book used for crafty is the general.ctg from Fritz5.32 ( which was said to be a collection of GM games similar to the crafty book ) and not a Kure book if I remember it right . 1.) But this isn't the main problem IMHO. Maybe we only see the Chessbase book learner in action (assumed the setting "optimal" is used ). Let's assume for a moment crafty is 100 points weaker than Fritz : so in general crafty would lose more than win against Fritz . The chessbase book learner in general seems to follow won lines and avoid lost lines in a very radical way . So Crafty loses in a Sicilian with black ? Bye-bye , Sicilian . Crafty loses a Ruy Lopez with black ? Bye-bye , Ruy . This is a little simplified but observed very often : so after a few games ( maybe 20 ) not only you begin to lose by a greater margin but what is even worse : you end up with a probably weaker book for your next matches . I like the original crafty way better which gives certain openings as set and only in this chosen area varies by learning and does the learning from the tips not the root . This isn't intented to be a scientific statement but it is interesting to watch the openings chosen when two Chessbase engines play each other and one is slightly stronger : seems it often goes this way then .
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.