Author: Robert Hyatt
Date: 06:07:21 03/19/98
Go up one level in this thread
On March 19, 1998 at 03:41:15, Ed Schröder wrote: >>Posted by Detlef Pordzik on March 18, 1998 at 19:44:04: > >>In Reply to: Re: SSDF and the programmers............ posted by Ed Schröder >>on March 18, 1998 at 04:38:38: > >>On March 18, 1998 at 04:38:38, Ed Schröder wrote: > >>>I fully agree with you. If things get honest again (engine versus >>>engine) >>>I will be happy to join the SSDF again. But let's face it, first we had >>>to deal with book cooks, then the learners came and now we have to deal >>>with an unknown autoplayer from a competitor? > >>Yes - this is right, let me say - the majority of it..... >>you know it, I know it - and all of them people of the elder times : >>what we now call book - cooking was named outbooking in them old times - >>but, it >>was more durable ( :-) ) - because of burnt EPROM's. >>This has allways been there..... >>because of lack of knowledge, how the learner actually works, I cannot >>comment this - you must know better, if there are more possibilitiers >>than just avoiding a loosing line, as a simple man'd suggest it to work. > >>>This for me is just the limit. Till now everybody has made his >>>autoplayer public. Now we have a new fashion, a secret autoplayer. >>>Who is next? Why should I join this new fashion? > >>Don't you do so, Ed ! > >You have my word. > > >>For my opinion neather you, nor Marty, Richard, Johann, Mark, Stefan and >>Robert >>would need to join such a " new " - but leading into nowhere - " fashion >>". >>( Jesus - hopefully didn't forget one of them big names....) >>This road'd be a dead end street - the basical quality of the progs is >>far too > >>high + in the end you all'd ruin the effort of many years. > >You are so right. > >We are now in a state that "improving learning" overrules the >importance of "improving the chess engine". All fine with me in >HUMAN-COMP games but not for COMP-COMP games on SSDF and (miss) >using AUTO232 for that purpose. > I have a big problem with this concept. IE commercial programmers make a big deal out of being #1 on the SSDF list. IE see your older Rebel web site, or now the ChessBase web site. And you all clearly imply that by being number one there, your program is the best of the available programs. OK, that's point #1. Point #2: Who are you selling your program to? To someone that is going to play it against other programs, or to someone who is going to play against it themself? The latter of course. So this person is going to use data that means one thing, and let it affect his purchasing decision for a program that he is going to use in a different way. Point #3: To do well against humans, you *must* learn. Most of the commercial programs that are run on the chess servers have "learning" even when they don't, because they are manually operated by a human that will either refuse to play someone twice, or refuse to let the program play the same opening again by forcing something else. So to do well in such an environment, you *must* learn or die. point #4: Crafty has "grown up" on ICC (mostly) although there are several hundred scattered around the other servers as well. It runs totally automatically. Without learning it could not function as well as it does, without continual book tweaking as I did for the first year or so of its development. It now "tweaks its own book" without any help from me which suits me just fine. :) I have not had time to make a "kit" for the SSDF guys, although Tony and others have run trial games for me helping to test the auto232 interface. But I have played hundreds of games against Rebel, Genius, Fritz, and other programs, without even knowing the outcomes unless I go back and look at the games mailed to me. But in each game, Crafty "learned" that an opening was good or bad. I have no idea if it has learned to cook some of your book lines or not. Probably has. not thru intentional actions on my part, but cooked nonetheless. If I play on SSDF one day, what does this mean? Did I intentionally bust your book? Do you play too narrow an opening selection? I don't have auto232 here. I wouldn't use it if I did because I play more games against GM players in one day than any of the commercial programs play in a year, and I value that kind of information more. Final point: I learned a long time ago, that the *only* person I can de- pend on is myself. If I get killed (as I did by WchessX) because someone has a clever learning algorithm that replays games I lose and avoids games where I win, then I can either complain about it, or take evasive action. I chose to take action, and now Dave and I have a good time watching Crafty and WchessX play and continually "probe" each others books learning what is good and bad. but the moral here is to fix it yourself, not to wait for someone else to change the rules to fix it, because you might wait for a long time. There are solutions to someone that has cooked up 100 lines... just make the learning more aggressive. My "result learning" simply disables a book move that leads to a loss (can be turned off if desired). I simply find the last point in the book where I had more than one choice and disable the move I played. I could speed this up by backing up a couple of choices as this entire line might be bad for me. I'll do it when I see a problem with someone booking up for me in this manner. But then I won't have to worry about outside help to solve the problem. >If I take the 100% Rebel9 chess engine (so no improvements at all!) >add the "learner" improvements as I have described in a previous >posting, and release this as Rebel_SSDF then Rebel_SSDF will end up >30-40 elo points higher on SSDF than Rebel9. > >Then I start "yelling" on the Rebel Home Page, "Rebel_SSDF is much >stronger than Rebel9!!". > >That would be a cheat to the public IMO. If this is true, then you ought to *never* yell about SSDF results, because it means they are worthless. > >In fact the only thing I would have done (please read my previous >posting) is that I have taken advantage of the fact that I know NOW >*HOW* SSDF testers do their testing. > >I don't want to be a part of such a development but this kind of >things is happening right before our eyes since a few years. > > >>Because then there would be no more research on increase of the quality >>of gameplay - but only the gasp for new horizons of tricks..... >>I fairly doubt - as I know some of you, that the people themselves would >>do it, after all. It would just be the upcoming end - correct ?? > >For me this is a correct conclusion. > >Others disagree. Others say see it as an innovation. Or as a >development you can't stop. Or as making a chess program as a >complete chess player. Or in a match everything is allowed. I fail to see "learning" as a problem. I see it as a natural evolution of the game. You took it up pretty late. But the most important point is that I doubt that learning as it is today will be how learning will look in 5 years. I have plans on learning "player by player" so that when IM Brian hartmann plays Crafty, Crafty will know which openings are most effective against him and spring them. Ditto for Roman or Shirov or Dlugy or any of the others that play games against me every day. So I'm probably going to be able to bust your book too, if you don't learn as well as I do. And if we only do learning like we do now, we are going to have the same sort of problem the Fritz guys saw in paris. Booking for one program is not the same as booking for another, any more than I can book for all humans at one time. I want to play tactically against positional humans, but i want to play strategically against tactical humans, because it makes them most uncomfortable. And that's what chess is all about, not just moving the pieces on the board... > >I say yes to all of this as long this is related to HUMAN-COMP but >not for COMP-COMP games on SSDF and (miss) using AUTO232 for that >purpose. > >I noticed that besides you and a few others most people do not >share my views I have tried to share. I guess I will lose this >discussion. So I like to be out. I hereby resign. > >- Ed - > Resigning is the easy way out. Don't do that. Fix Rebel. I'll tell you what I do as I do it. you can copy what you like, modify what you don't, and extend where you can. But if you fix it yourself, you don't have to depend on ChessBase releasing an auto interface any more than you have to depend on the SSDF for eliminating double games. Remember the three main actions here: (1) identify; (2) eradicate; (3) celebrate. :) > >>Dull ELVIS could offer a solution on short range, where the " loosing >>part " - >>in this case, SSDF, would have a minimum of problems - comparing to the >>state of art : >>just start a new testing seria with F5 - book on HD - why not....leave >>the brute >>searcher it's 44 MB - why not....R9 + M7 can grab 60 MEGS without >>problems - but >>in difference to F5 they don't NEED such big tablebases on 40/120...... >>repeat all this on commercial available autoplayer - and then replace >>the old result. >>IF F5 is still on top, then, it's allright - if not - I don't >>know...:-)) >>This would even leave CB the chance of not loosing their face ( totally >>). >>Not to forget Frans Morsch - who is really kicked by all this ! >>On longer range : >>ONLY commercial available progs with the original engine. >>Skip doubles. >>Standard autoplayer. > >>Maybe I'm a dreamer.....but I see nothin' better ; >>for the SSDF >>for buisness >>for the customer > >>>I do not share Ossie Weiner's opinions in the way he has expressed >>>himself. I have no single evidence the chessbase autoplayer cheats. >>>Neither do I expect that from a respected company. But I should have >>>the chance to check that myself. It's called fair competition. > >>As I've posted before - >>Ossi does his job - I think, you're one of the VERY last - not to know >>this, eh ? >>What does KK allways post standardly ? >>"....should be taken with a grain of salt...." >>- make the load a little bigger - then you can swallow this, too :-)) > >>>Sofar I noticed: > >>>#1. The chessbase autoplayer doesn't save the opponents game. Maybe >>>opponents in that stage update their learner? Logical place no? And >>>now maybe this learner update is bypassed? For Rebel8/9 this could >>>be so true. I can't check. How can I judge? > >>>#2. The chessbase autoplayer changes colors. white-black-white and >>>so on. This is not COMMON auto232. Looking at my source code this >>>doesn't seem to influence the learner of Rebel. But how can I know >>>for sure? And what about other chess programs? Did the Swedish ask >>>the programmers if this white-black-white behavior influence their >>>learners? One thing for sure, they didn't ask me. And from a >>>programmers view of point this white-black-white behavior can easily >>>disturb their learner. > >>>Quite a mess... > >>ELVIS ain't no programmer - I think, Robert wrote something interesting >>about this. >>Anyway - right in here ( without any elitary attitude ) - there are so >>many >>educated guys......names - older, newer - alot of them known for long >>time in this buisness - >>one should think about constructing an open letter ( why only Ossi....), >>maybe >>signed by 30 or 40 respectable people - so there'd be no chance to spot >>it as a single sided commercial sight - >>and then send it to Sweden......only a suggestion. > >>ELVIS
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.