Author: Vincent Diepeveen
Date: 07:50:46 05/27/02
Go up one level in this thread
On May 27, 2002 at 09:21:29, Uri Blass wrote: Please make a difference between position learning and booklearning. A program that's doing booklearning might be brilliant, but it will keep on losing the same game >On May 27, 2002 at 07:51:02, Vincent Diepeveen wrote: > >>On May 27, 2002 at 07:13:34, Jouni Uski wrote: >> >>>If they don't get Hiarcs8, why not to test Crafty18.15 with good book! Either >>>with Fritz7 book (if testing Chessbase native) or with WB2UCI converter and good >>>native book. Or is this waste of time? Second candidate is to test is may be >>>Yace. >>> >>>JOuni >> >>yes with fritz7 book would be a great test. But i am not sure they >>like in sweden crafty kicking butt of everything and everyone using >>the secret chessbase auto232 player. >> >>in fact they never post with which book who plays. > >I am really interested in the value of the Fritz7 book. > >It may be interesting if someone can test the same program against itself when >one side gets the Fritz7 book and the other side has no book and only use >learning to play a different first move after a loss. > >There are 20 different moves in the first move so the program with no book is >not going to lose the same game twice(it may also forget avoid moves after many >games because after 20 games with the same color it can be almost sure that the >opponent is different and there are only few cases when the ssdf do mathces of >more than 40 games and usually in these cases the number of games is 42 or 44). > >There is a possibility of losing the same game with different order of moves >twice(1.e4 e6 2.d4 and 1.d4 e6 2.e4) but if the learning is good enough then the >program is not going to allow a position that it knows that it lost and after >losing a game by 1.e4 e6 2.d4 d5 it is going to play a different move after 1.d4 >e6 2.e4. > >I guess that a program with no book and simple learning may be not more than 100 >elo weaker than the same program with the Fritz7 book. > >There may be even cases when the program with no book may repeat the same win >again and again. > >1.h3 gives equality and if a program wins by 1.h3 it may repeat it and if the >opponent has no reply in it's book against 1.h3 and no positional learning then >the opponent may lose again and again. > >Opening like 1.e4 a6 are also not a sure loss and I guess that the program is >not going to need to try some really bad moves like 1.e4 f6 or 1.e4 f5 or 1.e4 >b5 because it is not going to lose almost every game with black and these moves >are going to be the last in it's priority and it may try them only if it lost in >every other choice in the last 19 games. > >Uri
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.