Author: Uri Blass
Date: 11:03:18 11/25/00
Go up one level in this thread
On November 25, 2000 at 13:13:40, Thorsten Czub wrote: >On November 25, 2000 at 11:01:51, Uri Blass wrote: > >>I doubt how many humans will play Nf6 in a game. > >i meant the game itself is overall very human-like. >not only the Nf6-move. > >>I believe that most humans are going to play Ng3. > >could be, yes. >what about tal ? :-)) Tal is an exception and most humans do not play in his style. Saying that > >>If white can win by another move instead of Nf6+ then Nf6+ is a mistake and when >>the opponents will get better playing this move is not going to lead to a win >>against them. > > >right. accurately spoken. > >>I believe that the evaluation after search should be accurate. > >jajajaja. verstehe. sure. accuracy. trying to be god. > >the tarrasch-syndrome. >playing like a machine (capablanca) = without mistakes. > >IMO your target is a wrong target. > >even spock has learned this. > >>My opinion is that not accurate static evaluation can be a good idea only if it >>helps to get more accurate evaluation after search. > >there is no accuracy. this is a concept that only exist in your mind. >accuracy only within a range. and since chess is not outsearched, >the range is limited. I agree that you cannot be accurate always but it is possible to be more accurate. There is a problem what is the definition of being more accurate. The only right evaluation is the evaluation that give only 0,1/2,1 but you cannot practically achieve this evaluation. I believe that the evaluation should try to predict correctly the result of the game(assuming that both players have the same strength). It is possible to get a scientific definition for being more correct in predicting the result of the game but it is not simple and I am not going to discuss it now. > >>When I play correspondence games I expect my opponents to be accurate > >aha. >and what about your wife ?! > > >>and if I >>find that Nf6+ is leading to a draw then I play another move that gives me a >>better chances. > >this was 60/60. >it was a 20 games match. >when you have a time control, you cannot be accurate all the way. > >>I disagree. > >of course you do. > >>If there are 2 moves that are leading to a draw an accurate program can choose >>one of them that is a sacrifice. > >an accurate program would rarely play a move that is a sac. >also it would see that the move can be defended. so why should it expect >to play the move or play it itself when it also sees a defense. It depends on the definition of an accurate program. Junior can sacrifice pawns if it believe that the positional value of the attack is bigger. Trying to do the evaluation more accurate is not trying to avoid sacrifices. In the case of sacrifice that lead to a draw it is possible that no other move wins and if the evaluation of the sacrifice is not worse than other moves it can choose it. > >>I think that no program is accurate and that no program is going to be accurate. > >aha. but you think one should try to be accurate ? Yes. I think that trying to be more accurate can help programs. > > >>If crafty is going to play accurate then it is never going to lose in chess. >>The fact that it is losing is a proof that crafty has no accurate evaluation. > >aha. >so what means accuracy ? >you want to play the best move in a position. > >but what is best ? >the winning move ? >what do you do when there is no obvious winning move. >what is best then ?! >there is no move to mate in x and no move to win material in x. > >what is best move in x then ? > >>I disagree. > >thought so. > >>There are humans who never sacrifice in chess and I do not call them ill. > >you misunderstand me. >i said not : never saccing. i said never doing anything because you saw >in forward that it makes no sense. >that it loses. >so better doing nothing. > > >>Crafty is not accurate and there are a lot of cases when it evaluates wrong and >>this is one of the reasons that Crafty lose games(it can sacrifice the king >>safety for a pawn because of wrong evaluation function that say that the pawn >>has bigger value). >>Gambittiger is also not accurate and can do the opposite mistake. > >right. not accurate. >i will send back the cd. i want an accurate cd ! especially for mail-chess. >accurate. accurate. i want more accuracy. piep. I am not going to send back the CD. I did not expect accurate evaluation when I bought gambit. I know that there is no program and no humans with accurate evaluation. Gambit is one of the programs that I am going to use for my games but I am not going to use the scores of it and the main reason is the fact that it is a preprocessor and I cannot compare between scores of different nodes in the tree. There are programs without this behaviour(Gandalf) Gandalf can be wrong in the evaluation but at least it is wrong in the same way and I have not the noise of being wrong in a different way because of the root position. 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.