Author: martin fierz
Date: 01:18:43 10/10/02
Go up one level in this thread
On October 09, 2002 at 23:38:02, allan johnson wrote: >On October 09, 2002 at 19:36:58, martin fierz wrote: > >>On October 09, 2002 at 18:29:18, Ed Schröder wrote: >> >>>On October 09, 2002 at 17:40:34, Thorsten Czub wrote: >>> >>>>its not that we expected fritz to win, but we expected fritz >>>>to stand, and maybe to get a few chances. >>>> >>>>to die in pride. but not to die without any chance. >>> >>>I remember the first 3 games against GM v/d Wiel where John completely outplayed >>>Rebel resulting in a 1-2 score where it should have been 0-3. Then after the >>>weekend the situation turned completely in favor of Rebel and the next 3 games >>>ended in 2.5-0.5 (where it should have been 3-0) in favor of Rebel. >>> >>>Bottom line: to early yet for an end conclusion. >>> >>>Ed >> >>hey ed, since you have "been there, done that": was this comeback of rebel due >>to "chance" (i.e. just a statistical fluke that vdW played well first half, >>rebel 2nd half), or did you actively intervene and change something about rebel >>to turn the match round? and if yes, would you mind sharing with us what you >>did? and do you have any ideas what the DF team would have to change if they >>were allowed to? >> >>aloha >> martin > martin: Correct me if I'm wrong but didn't Kramnick get the opportunity to > familiarise himself with Deep Fritz? right. > If this was the case then perhaps the > surprise element could come into play especially in the short term. i just wonder how much you can really change in a program with a little bit of tweaking. i guess not that much. probably all you do is change the weights of a few eval terms. my guess is that the problem of the programs is not in the weights of their eval terms but rather in the missing features in the eval. so fiddling around with the parameters probably wouldnt change very much. specially since kramnik doesnt seem to focus on any particular weakness of fritz so far, but just on the general fact that fritz is weaker in the endgame than he is. and that is not something you can make go away with a few slight tweaks in the eval IMO. >If he was > allowed to do this how was it justified? kramnik dictated the match conditions. in this sense, it was just "a given", and there is no need to justify it. kramnik just demanded it and chessbase gave in, because they wanted this match. if you were to ask me for a justification i would say that the programmers have been preparing for these matches for YEARS and have special code in their programs aimed at the weaknesses of humans thanks to this experience (e.g. some programs score a position with all pawns on the board as bad for themselves, because the programmers know that in closed positions humans are better. this kind of eval obviously has nothing to do with chess, but is just aimed at opening the position for play against humans). so kramnik wanted to catch up in this respect. i believe that a copy of fritz 6 would also have done the trick - the programs dont change that quickly, and kramnik would probably have worked out the same strategy as he has now. chessbase had the right to change the opening book completely, and IMO they should have done that and i think they did. which should make specific preparation of kramnik useless, but the general ideas which he developed (i.e. get the computer out of book with sensible moves and into queenless middlegames/endgames) still would apply. aloha martin
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.