Author: blass uri
Date: 23:16:40 04/20/00
Go up one level in this thread
On April 20, 2000 at 10:45:41, Amir Ban wrote: >On April 20, 2000 at 05:34:28, blass uri wrote: > >>I found that Junior6 tried to win by repeating the same opening against >>Rebel8(p200) without success because it chose a different move in the middle >>game. >>I suggested Amir a long time ago what to do against this problem(learning to >>have winning game as part of the opening book) but he does not use my idea. >> >>Another mistake of Junior is resigning in lost positions(in part of the cases) >>I saw cases when Fritz failed to win a simple winning position because of >>tablebase bugs and if the ssdf list was important for Amir he could tell Junior >>not to resign. >> >>I think that the ssdf rating list is not the important thing because we have no >>idea about the size of error in the not public games and the important thing is >>the public games in the ssdf list. >> > >The SSDF rating list is important for me, but I have limited time to develop >Junior and would not waste it on this kind of "improvements". > >To make clear, I think book learning, however aggressive, is a legitimate >offensive and defensive tool, but "learning" of entire games that an existing >program is guaranteed to follow I consider to be primitive and having nothing to >do with the real competition between programs or programmers. Note that one >programmer who was reported to be engaged in this practice has now fallen behind >and dropped out of the competition. I do not see a difference between repeating the same opening and repeating the same game. I found that Fritz5(p200) could repeat the same game against Rebel8 with slower hardware 5 times when Junior6 failed to do the same against Rebel8(p200). Learning to repeat the same opening without learning to repeat the same game gives an advantage for deterministic programs. I understand that people are more interested in the power of the engine but I do not see why repeating the same opening is right when repeating the same game is wrong. > >It is said that no game has ever been won by resigning, and this is true. On the >other hand I've never heard of a chess player or program whose strengths include >refusing to resign on time. I agree that it cannot give a lot of ssdf elo rating but I guess that refusing to resign can improve Junior's rating by about 5 ssdf elo rating. possible reasons: 1)bugs of programs 2)cases when the evaluation of Junior is wrong You know about a case when Junior evaluates a draw position as a mate against itself because of stalemate identification problem. In this case Junior does the wrong move but I can imagine a case when Junior resigns instead of doing the right drawing move. Not resigning in a clearly lost positions against humans is a bad behaviour but against programs that are not operated manually I do not think that there is something wrong with it. 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.