Author: Bertil Eklund
Date: 16:21:25 12/26/98
Go up one level in this thread
On December 26, 1998 at 18:04:32, blass uri wrote: > >On December 26, 1998 at 15:50:10, Christophe Theron wrote: > >>On December 26, 1998 at 05:47:46, Bertil Eklund wrote: >> >>>Fritz3 have learning, if you play with big hash-tables, check there is a file >>>called Fritz3.lrn it works ok. It´s not book-learning but it works try to win a >>>game in the same way and Fritz3 tries another move in the place the score >>>dropped significantly. >> >>Do you know if this learning feature works even if the game is not saved? >> >> >> Christophe > >I found that game 5 and game 9 of the match Junior5-Fritz3 were basically the >same(The first different move was move 37 of Fritz3 (the evaluation of Junior5 >was 9.75 pawns advantage for Junior5 at move 37) > >Is it possible if I assume that the learning feature of Fritz3 works? > >Uri The problem is well known, where should the book learning play another move, Often the new move is played much to late and don´t matter. A simple learner can for instance recognize a mistake when the score drops i.e. 1pawn but it is not allowed to play a move that is calculated to more then a half pawn worse then the earlier played move, the consequences can be as in the game you mentioned.The learners of today McPro, Hiarcs, Genius5 is clearly improved over Fritz3. Bertil 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.