Author: Bertil Eklund
Date: 14:33:37 11/19/05
Go up one level in this thread
On November 19, 2005 at 05:50:31, Uri Blass wrote: >I wonder what happened to the famous learning of the chessbase programs > >From the fruit page about future plans: > >"Book-learning and position-learning (perhaps but not very likely - >book-learning is not easy to implement in a UCI-Engine)" > >Note that the chessbase programs are supposed to have book learning and I >remember that old Fritz beated Rebel8 again and again in the same line because >Rebel did not have book learning. > >People could expect based on these observations that Fruit is going to lose >matches in the ssdf because of lack of learning (or at least to score worse in >the second half of the match but I do not see it happens). > >I wonder if there is a bug in the implementation of learning by chessbase. > >Note that based on the match against shredder9 that is supposed to have >positional learning it seems that positional learning also did not help Shredder >against fruit. > >I did not look at the games of the ssdf to find out but I wonder if the >opponents of fruit did not try to repeat lines that they beated fruit and I >wonder if it had no effect(even if they cannot repeat the same lines exactly I >expect repeating similiar lines to have bigger probability to lead to similiar >results). > >Is there something wrong in my assumptions? > >Uri Positive learning don't work with (if not the opening are totally winning) engines like Shredder, Nimzo and Comet for example, because these engines plays different moves and lines (if not forced) in each and every game because of hashing (?) or some other factor that makes them play different moves in the same position. Bertil
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.