Author: Matthew Hull
Date: 08:07:45 02/27/04
Go up one level in this thread
On February 27, 2004 at 10:18:15, martin fierz wrote: >On February 27, 2004 at 09:08:51, Matthew Hull wrote: > >>On February 27, 2004 at 05:46:13, martin fierz wrote: >> >>>On February 26, 2004 at 23:08:28, Robert Hyatt wrote: >>> >>>>On February 26, 2004 at 17:42:43, martin fierz wrote: >>>> >>>>>On February 26, 2004 at 14:59:55, Robert Hyatt wrote: >>>>> >>>>>>On February 26, 2004 at 04:37:37, martin fierz wrote: >>>>>> >>>>>>>On February 25, 2004 at 22:42:08, Robert Hyatt wrote: >>>>>>> >>>>>>>>On February 25, 2004 at 18:40:06, Bas Hamstra wrote: >>>>>>>> >>>>>>>>>On February 25, 2004 at 13:46:22, Robert Hyatt wrote: >>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>What book are you using for Crafty? >>>>>>>>>> >>>>>>>>>>It will _never_ play 1. g3 as white with any book I have ever distributed. Not >>>>>>>>>>that it is a bad move, but it suggests that something is way wrong with the >>>>>>>>>>setup you are using for Crafty, at least. >>>>>>>>>> >>>>>>>>>>Aha. I see you are using the fritz powerbook with max variety... >>>>>>>>>> >>>>>>>>>>That begs the question of what your "tournament" is supposed to show, since a >>>>>>>>>>wide book introduces _lots_ of luck into the outcome, and won't be reproducible >>>>>>>>>>by anyone else since nobody uses one common book for multiple engines... >>>>>>>>> >>>>>>>>>What do you mean "nobody"? Everybody does do it all the time! And a wide book is >>>>>>>>>even better, it shows what your search-object (engine) is capable of in a wide >>>>>>>>>variety of positions, in stead of playing over and over and over the same few >>>>>>>>>"proven" openings. If Crafty is mated in 12 moves in an irregular opening, >>>>>>>>>wouldnt' that be interesting to know? Think about it. >>>>>>>> >>>>>>>>Not if that opening is 1. g3, which neither it nor I (nor most anybody) will >>>>>>>>play. Ditto for the 1. f4 openings, the 1. b4 openings, etc. >>>>>>>> >>>>>>>>I don't write code to handle such cases, if I never expect to have to play them >>>>>>>>over the board... >>>>>>>> >>>>>>>>Now if you choose _reasonable_ openings, that might be another matter. But I >>>>>>>>don't particularly like 1. g3 and after having played chess for 40+ years as a >>>>>>>>human, I _still_ don't ever play that opening... >>>>>>>>> >>>>>>>>> >>>>>>>>>Best regards, >>>>>>>>>Bas. >>>>>>> >>>>>>>i think bas is very much right. i test with nunn2 positions. these cover a lot >>>>>>>of variety, closed and open positions, positions with opposite-side castling >>>>>>>etc. you get a better idea what an engine can do if you test lots of different >>>>>>>positions, and also what it can't do! >>>>>> >>>>>>And once you learn that it can't do something (say a hyper-modern type opening) >>>>>>very well, what then? I just say "don't play that opening" and go on, and maybe >>>>>>when I have time, at some point in the future, I might address that. IE it was >>>>>>a long time before I would let Crafty play any fianchetto sort of opening as it >>>>>>didn't understand how critical the bishop is to defend the weak squares caused >>>>>>by the g3/g6/b3/b6 pawn push. Once I fixed it, I allowed those openings. But >>>>>>until I did, I did not. I would call it silly to make an old program of mine >>>>>>play such openings, because I _already_ know that it will do badly with them. >>>>>>What is the point of seeing that again? >>>>> >>>>>that's not the point. if crafty can't handle a position with the fianchetto, >>>>>then odds are it probably doesn't understand how to play against it either... >>>> >>>>Actually, it plays _against_ it very well. It just does the usual "sieze the >>>>center and break it open." But once you have played g3, you don't really want >>>>to see any e4 type stuff as it creates weaknesses, so the "occupy the center >>>>with pawns" and such is wrong. >>>>> >>>>> >>>>>>>i don't understand your take on the opening moves. 1. g3 is a sound move. 1. b4 >>>>>>>is slightly weird, and 1. f4 is really weird. >>>>>> >>>>>>I think f4 is pretty good, in fact. >>>>> >>>>>once again, that is where our chess rating difference comes from :-) >>>> >>>>Not sure what you mean. Should I point you to some GM games with 1. f4? :) >>>>I have several thousand. >>> >>>fine, you can believe what you want of course. you can also try to count the >>>number of games that kasparov, karpov, kramnik, anand & co start with 1.f4. or >>>the number of games played in world championships starting with 1.f4. or go and >>>ask some of your GM friends what they think of 1.f4... >>> >>> >>>>But I am talking about _computer_ games. And against human opponents, 1. f4 is >>>>really not a bad opening at all, as white's f4 move is an aggressive move in >>>>many cases from the Sicilian to the King's gambit.. >>> >>>err, now you are changing the subject rather radically. we're talking about >>>1.f4, which is a crappy opening. playing f4 at a later stage, when the position >>>is completely different is a rather different matter :-) >>> >>>of course, if you play 1.f4 with crafty against humans then i would say it's >>>fine if you get a king's gambit e.g. if the opponent replies to 1.f4 with e5 and >>>you go 2.e4. but why not start out with 1. e4 and play king's gambit against e5 >>>than? >>> >>> >>>>>>But while g3 is perfectly sound, white is >>>>>>saying "I am going to play on the wings in many variations (while in others you >>>>>>might see a quick d4/e3/etc of course). And my program simply doesn't like that >>>>>>idea very much. >>>>> >>>>>same answer as above: if you can't play one side of an opening, you probably >>>>>can't play the other either. >>>> >>>>I totally disagree. There is a difference between playing defensively, and >>>>playing to attack a defensive player. Ditto for a player that eschews the >>>>center. I don't have to play such openings well, in fact I don't even have to >>>>like them at all, in order to play against them from the other side... >>> >>>i totally disagree too. if not having the center is such a problem for crafty, >>>it will evaluate all positions against such a passive opening as being just >>>great for black, when in fact they are not - they will be about equal. having a >>>way-off eval for these positions can give strange results; e.g. you will get >>>into bad positions thinking they are good for you because you have a bit more >>>central control. >>> >>> >>>> >>>>> >>>>> >>>>>> >>>>>>>i also wouldn't want to test too >>>>>>>much with moves like 1. f4 or 1. g4; but 1. g3 is fine - it mostly transposes to >>>>>>>some regular opening with fianchetto like some form of catalan or english, which >>>>>>>are good, solid openings. it definitely qualifies as reasonable! >>>>>> >>>>>> >>>>>>Depends on your definition of "reasonable". "sound opening"? Yes. But >>>>>>"sound opening for a program that doesn't like the resulting positions at the >>>>>>moment?" No. >>>>>> >>>>>> >>>>>> >>>>>> >>>>>>> >>>>>>>finally, i have played openings that are slightly unsound just to learn about >>>>>>>the resulting positions (e.g. queens gambit tarrasch defence to learn about >>>>>>>IQP). you can stop playing them again later, but you will have learned something >>>>>>>you can apply in similar positions arising from other openings. for engines the >>>>>>>same: if crafty cannot handle positions that come up after 1.g3, there is some >>>>>>>kind of problem in crafty.... >>>>>> >>>>>>Perhaps the problem is already known? And discovering it a second, third or >>>>>>fourth time is not exactly going to reveal anything new... That was my point. >>>>>>That is why I release books with my engine. I consider a chess program to be a >>>>>>combination of engine, book, endgame tables, configuration files, and the like. >>>>>>Change any one of them and the "program" is now "different". IE I'll play you >>>>>>as many games as you want (human to human) but if you ask me to play 1. g3 I'm >>>>>>not going to comply. I have other openings I like far better. :) That is the >>>>>>idea here, IMHO. It makes no sense to force the program to play something it >>>>>>doesn't "like". >>>>> >>>>>you are thinking of maximizing playing strength only. what if i wanted to use >>>>>crafty as analysis module in chessbase? i would want it to make reasonable >>>>>suggestions in all openings! i never use chess engines to play against. i always >>>>>use them to analyze. >>>> >>>> >>>>I am only considering playing computers against opponents. Nothing more or >>>>less. So you are correct in that regard... But when tournament results are >>>>reported here, do you really think people say "hey, that program did pretty well >>>>playing an odd variety of openings" or "hey, that program got beat pretty >>>>badly"??? :) >>> >>>no, of course they will say the second. i just don't like the entire approach. >>>there are many cop-outs i know of in computer chess, like all those tricks with >>>thinking "many pawns on the board = bad position for me", or your trick of not >>>taking something on g5 even if it's for free. or the famous "fritz won't take >>>free pawns on e4 because of mr. nemeth" thing. these are the extremes. this >>>example is less extreme, of course. but still: chess is chess, and programs that >>>don't know how to handle certain positions will always be susceptible to some >>>form of attack. take kasparov-X3d fritz, game 3. most chess programmers try to >>>solve the problem of closed positions by hoping their opening book will stop >>>them from getting into such positions. and when it happens, BOOM, there goes >>>your 2800 rating and even 1800 is still too much... i'd rather try to fix the >>>problems in my eval than continuously fix the opening book. >>> >>> >>> >>> >>>> >>>> >>>> >>>> >>>>> >>>>> >>>>>>Do you think you could coax anything but 1. d4 out of (say) Korchnoi, when the >>>>>>game is important?? >>>>> >>>>>certainly. Nf3 and c4 at the very least. modern top grandmasters play >>>>>everything, every single one of them. there are very few GMs who stick to a very >>>>>narrow repertoire (eg sveshnikov, lputian), and none of them is in the very top. >>>>>coincidence? >>>> >>>>Missing the point. Do they play _random_ openings? Or do they play openings >>>>they have studied and prepared at home, often for specific opponents. Hint: It >>>>is _not_ random... >>> >>>hint: crafty has not studied any openings. >> >> >>GM's don't play random openings well. They play their repetoire well. Same >>with programs. > >most programs have similar weaknesses. e.g. they don't play closed positions >well. if you remove the well-adjusted finetuned book from all engines, as is the >case in this tournament, you are not putting any of them at a serious >disadvantage. except if one of these engines is really much more helpless in >many classes of positions than the others. and if that is the case, then i am >glad to hear about it! > >>>all other programs participating in >>>this tournament have not studied any openings. >> >> >>The programmers have studied them. > >really? show me a 1600-rated programmer who has studied an opening seriously :) >they don't study openings. they notice their engine has no clue, then they >remove it from the opening book. that's not studying - that's exactly the >opposite! instead of facing the problem, they brush it under the carpet and hope >it doesn't show up again. and if/when it does, then it's not their fault... > >>The goal is to build a system that will win, >>not to build a system that will win with inferior openings. > >depends on what you want. i use chess engines for analysis. there the engine has >to deal with whatever i throw at it, not with what is in it's own book. if it >can't handle some normal chess positions because the programmer is only masking >instead of addressing the problems of his engine, then that engine is useless to >me. look at kasparov vs X3D fritz game 3 to see what i mean... > >blaming opening books for bad results is a common practice. that doesn't make it >a better practice though! GM's play what they are good at. Same with progs. Your point would seem to be that you expect a prog to be all GMs rolled into one so you can analyze. OK. That's really impossible until the game is solved. Until then, a programmer will focus on competing, which means building a book+program combination that wins. > >cheers > martin > >> >> >> >>>i'm pretty sure that if another >>>strong program was at the bottom of the standings, it's programmer would use the >>>same excuse :-) >>>i'd say that 30 games are not enough to be significant and leave it at that... >>> >>>cheers >>> martin >>> >>> >>> >>> >>>>>cheers >>>>> martin >>>>> >>>>>> >>>>>> >>>>>>> >>>>>>>cheers >>>>>>> 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.