Author: Robert Hyatt
Date: 19:48:19 10/10/02
Go up one level in this thread
On October 10, 2002 at 21:06:11, martin fierz wrote: >On October 10, 2002 at 19:58:14, Robert Hyatt wrote: > >>On October 10, 2002 at 19:19:07, Wayne Lowrance wrote: >> >>>On October 10, 2002 at 18:26:28, Robert Hyatt wrote: >>> >>>>On October 10, 2002 at 18:15:48, Jorge Pichard wrote: >>>> >>>>>On October 10, 2002 at 18:06:12, Robert Hyatt wrote: >>>>> >>>>>>On October 10, 2002 at 15:00:33, Jorge Pichard wrote: >>>>>> >>>>>>>On October 10, 2002 at 11:25:53, Graham Laight wrote: >>>>>>> >>>>>>>>Considering that Fritz lost two of the first four games, and drew the other two, >>>>>>>>winning three of the remaining four games (the minimum needed to win the match) >>>>>>>>is a tall order. >>>>>>>> >>>>>>>>-g >>>>>>> >>>>>>>At this point even one more draw and a win is better than what most of us here >>>>>>>expected from Deep Fritz. I still have hope for DF to win the first game of the >>>>>>>match, and as Kramnik becomes more fatigue the better the chances for Deep Fritz >>>>>>>to win one single game. This is a great accomplishment for Deep Fritz team to be >>>>>>>able to draw at least 3 games, considering that Kramnik had the actual program >>>>>>>before the match. This performance is much better than Deeper Blue of 1997, >>>>>>>considering the condictions set before the match for Kasparov. >>>>>>> >>>>>>>Pichard. >>>>>> >>>>>> >>>>>>No way DF could surpass the result posted by DB in 1997. It might _equal_ that >>>>>>result if >>>>>>it manages to win. But not if it gets drubbed as it has in the first four >>>>>>games... >>>>> >>>>>Kasparov would have also drubbed Deeper blue if he was allowed to play against >>>>>deeper Blue for at least a month before the match started. Lets NOT just count >>>>>the # of Win or Draw here, lets consider the case under the same circunstance. >>>>> >>>>>Pichard. >>>> >>>> >>>>Based on what? The match rules would _still_ have allowed modifications to be >>>>made between rounds. But Kasparov couldn't possibly have played against Deep >>>>Blue >>>>until it was completed, which happened a week or so prior to the match. >>> >>>I have stayed out of this DB thing. But I would have to guess that DB would not >>>have won the match if it was held at a later date with the same rules that DF >>>has to live with. Just conjecture of course, but this whole match has pissed me >>>off from the very start because of the rules, unfair I think >>> >>>Wayne >> >> >>This is a classic case of "self-screwing". :) The conditions are ridiculous, >>and the >>match can be called the same now, since we _knew_ how it was going to turn out, >>given the conditions that are in effect. > >i don't believe a word of what you are saying. except if the DF opening book in >the match is the same as the one as kramnik got, under the rules. that would >make a big difference. else, the DF team just screwed itself by not changing >it's opening book, or not enough, for this match. and until shown otherwise, i >don't believe that kramnik worked out any of these games in any detail. I don't believe he worked out _any_ "game" in detail. I _do_ believe he played a lot of games and picked up on key weaknesses that could be repeated, by reaching similar (but not identical) positions. I've seen this happen too many times on ICC to count. > he is >playing his standard openings. he is not playing weird moves like 1. a3 to get >into some worked out game. I never said he was doing that, and hope I never implied I thought he was. He simply has found weaknesses that lead to predictable play in particular types of positions, whatever they may be. The rules prevent _any_ changes to the program excepting for the opening book, and changing the opening book would only be an attempt at dodging around a problem rather than doing something to change the basic way the program behaves... >since kramnik is playing what he usually plays, the DF team should have >anticipated that and changed the book for kramnik's main lines. if they didnt, >shame on them. I don't think the book is the issue... > >in response to your other post: > >>I have thought more about this, and I would make the following >>controversial statement >>(the reasons will eliminate the controversy however): >> >> Crafty _would_ do better than Fritz in this match. Not that I would >>expect it to win, >>but it _would_ do better. > >this is indeed controversial! and your reasons don't convince me: > >>Now for the "why?" >> >>1. If we stick crafty in _right now_ all the weaknesses he has found >>in Fritz are useless. > >"all"? you must be kidding! his general strategy is to go into relatively simple >endgames. which is a good idea against ANY program - be it fritz or crafty. Not so easy. Crafty does handle some endgame positions as well or better than a GM. This comment comes _from_ GM players that have both played it hundreds (or thousands) of games as well as from a couple of GM players that give me advice from time to time. Yes it has weaknesses. But _not_ the same weaknesses as Fritz. It understands a lot about majorities, candidates, pawns on wings, etc... Enough that it would take him time to discover what it didn't understand, and he only has _eight_ games to do that... Not enough with me tweaking between rounds to keep things off-balance. > >>Crafty evaluates endgames differently >>(and better in many but not all cases). > >emm, "hype"? how do you know that crafty evaluates endgames better than fritz? Because I have watched _many_ endgames played between the two. And you can readily see the kinds of things Fritz does well and the kinds of things it does poorly. IE it advances passers quite well. But it also voluntarily trades into endings it simply can't win. Once it has started the unavoidable series of trades, it begins to "see" because the search exposes part of the problem. But it has to recognize that if it has (for example) two connected passers in the middle, and I have majorities on both wings, that trading to a king and pawn ending _loses_, not _wins_. That is what I base my comment on. Actual games played. And note I am not saying Fritz is a bad program. It will win more vs mine than it loses. But not so many in endgame positions, overall. >do you have anything to base this claim on? fritz won this rook endgame test of >sune... not that rook endings are all there is of course, but i don't really >think fritz is weak in endings. > >> So all his pre-match >>preparations would go right down the toilet, and the onus would be >>on him to find Crafty's (many) weaknesses, OTB in games that count. >>Not that easy. >i'll claim that one weakness in crafty is exactly the same as DF: go ahead and >set up the position of game 2 after the moves 34...Kd7 35. Rc5. tell me that >crafty would play 35...Rc8! and i will shut up. AFAIK ALL programs will rather >go for a passive defence in a bad rook ending and be slaughtered rather than >giving a pawn for counterplay. if crafty does not see ...Rc8!, it has the same >weakness as fritz in this rook endgame. Does it matter? Kramnik said he was winning well before that... I don't see a reason to want a program to prefer a less losing move over a more losing move. It's more important to avoid the losing position from the beginning... > >>One mistake is all he has to make to lose a game, >right. but he has not even come close yet in his games against fritz. whenever a >mistake by kramnik comes, it will be exploited by fritz just as well as by >crafty or any other top program. The problem is, he has found a "style" of play that minimizes his chances for errors, because he has found that without queens, a program that relies on aggressive play suddenly becomes clueless when the attacking chances are not present. And he _knows_ that before he starts the match. Were this my program, he would not because I simply would not have agreed to such a ridiculous set of rules. Of course, he would never play Crafty anyway because he wanted those rules and I would never have agreed to them, so end of match before it could get started... > >>and probing for unknown weaknesses would not be >>easy with only eight games to play, and with the program possibly >>"changing personalities" between rounds. >changing personalities means changing eval weights, right? sounds good, but most >of kramnik's knowledge of chess has eval weight 0 in fritz and in crafty too. >changing the weights of all your parameters still leaves that 0 right there... Not even close. Kasparov complained because deep blue seemed to "change". In one game it seemed to evaluate bishops too high, in the next, not high enough. That is enough to disrupt a plan, and if you only have 8 games to develop a plan, I believe he would have a _much_ harder time. Yes, I believe he would win. He might even win with the same margin of victory. But he would definitely have to "work" for the victories, rather than using pre-obtained knowledge to steer the program into never-never land with little chance of anything bad happening at all. > >>2. If Crafty had started this match it would have done better. >>Because I would _not_ have agreed to the silly rule about "he gets a >>copy of the program beforehand and no changes can be made to the >>program after that unless he gets the new version far enough in advance to >>test it thoroughly." I have no doubt Kramnik would still win. >>Perhaps he would even win every game. But the games would _not_ be >> boring endgame losses, the endgames would have inbalances that >>would make them quite interesting, as would the middlegame. >and what makes you think that? why would the endgames be different with crafty? Because you can't simply trade into a pawn ending with Crafty and expect to roll it over. It has beaten Shirov in endgames _many_ times. Even in opposite-bishop endings. Because it does have some specific code for some of those cases... Crafty has "grown up" trying to thwart the GM goal of finding a hole and sitting on it for weeks... Or with the "drawmasters" just playing for draw after draw. Read some of the finger notes on ICC. Some "commercial" program operators won't play IM/GM players that specifically try to play for a draw, over and over, because they succeed 90% of the time and kill the rating. I have chosen to combat this in a different way, with some success, although there is plenty of room for imporvement. But one big point... 5 years ago, with Crafty or Cray Blitz, I hated to see a GM trying to reach an endgame because the programs (both mine) would have serious problems. With some help from Roman, I don't fear endgames against GM players _period_. Yes it will lose some, but it will also win some just as well. The GM comments show that some of their advice has helped a lot.. > >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.