Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / The Trilogy is complete!
- - By lkaufman (*****) Date 2008-08-10 02:49
     In the last few weeks, Rybka has played three matches of four games each at intermediate time limits (30'+30" twice, 30'+20" once) with titled players. I had made it a goal to be able to defeat a GM at pawn (f7) and move, an IM at pawn and two moves, and an FM at pawn and three moves, and Rybka has now met the goal in all three categories, with a total score of 9-3. Each of the three players had at one time been near the top of their respective ranks: "Dzindzi" was in the World top ten, Eugene Meyer was just one norm away from the GM title, and John Meyer was close to IM strength when he tied for first in the U.S. Open. Even if they are all well past their prime, I think they can each be considered representative of their respective ranks.
      We have yet to prove that we could win such matches at a FIDE standard time control, but in view of the decisive overall margin it seems likely. In any case we can now confidently recommend these handicaps for serious training by players in each of these classes. So if you are able to win a four game match at 30'+30" with Rybka 3 on strong hardware (let's say at least a 64 bit quad) at one of these handicaps, you have every reason to think of yourself as being competitive with FM (pawn and 3), IM (pawn and 2), or GM (pawn and move) level players, even if you lack the title or required rating. If you are playing on lesser hardware, you need to advance one handicap step more before making this claim. I'll be interested to hear from any forum readers who can win a match at a handicap that equates to less than their actual rank (for example, an untitled player who can beat Rybka 3 on good hardware at pawn and three or on ordinary hardware at pawn and two).
Parent - - By exigentsky (***) Date 2008-08-10 05:09 Edited 2008-08-10 05:35
I wouldn't extrapolate too far from this. The sample size is small and the conditions are not normal chess. Performing with humans at whatever level is still the only real measure. I suppose "Father" may be a GM by a lot of these standards (while the time controls were faster, he didn't even have a handicap).
Parent - - By lkaufman (*****) Date 2008-08-10 05:20
True, this is just an indication that a player might have the ability to earn the corresponding title if he learns openings properly, etc. As for "Father", first of all I haven't heard of him playing Rybka 3 yet, and anyway getting occasional draws does not meet my criteria of winning a full match at the proper handicap.
Parent - - By exigentsky (***) Date 2008-08-10 05:33
"might have the ability to earn the corresponding title if he learns openings properly," "Might" is the operative word there. So far, we just don't know. However, it is still a good challenge.

BTW: Does 30'+30" mean 30 minutes and a 30 second increment?
Parent - By lkaufman (*****) Date 2008-08-10 15:33
Yes, don't they teach those ' and " symbols in the schools any more?
Parent - - By Vempele (Silver) Date 2008-08-10 08:27

> As for "Father", first of all I haven't heard of him playing Rybka 3 yet

Evidently you don't read Talkchess.
Parent - - By lkaufman (*****) Date 2008-08-10 15:25
No I don't; I don't even have time to read everything on this forum. Regarding the one draw "Father" reports, there are two countermeasures available for his tactics. First of all a high contempt setting is appropriate against a human; I would recommend 50 for a grandmaster and perhaps 80 or even 100 for a strong amateur player like "Father". This is quite helpful in foiling his tactics. Also I think that the "Human" version would be more effective than the default against his play. Finally, the right opening book may help; I don't know whether the Convekta book or Jeroen's Rybka 3 book works better against these drawing tactics.
Parent - - By Fulcrum2000 (****) Date 2008-08-10 15:48
Father has a really hard time with R3 :

Father (To all):  but finally Rybka 3 is a monster eating walls

Father (To all):  this is Rybka 3
Father (To all):  I am going to work hard fot develoment of better chances
Father (To all):  people invited
Father (To all):  man vers machine
Father (To all):  first summary
Father (To all):  Rybka 3 broke the wall ever in b5 to black and b4 to white
Father (To all):  then some new must be invented
Father (To all):  I am working in that...

Parent - By lkaufman (*****) Date 2008-08-10 15:56
I'm glad to hear this. I made some changes just a few days before Rybka 3 was set in stone that were intended to help this, even though they might have cost one Elo point or so in engine vs. engine testing. Looks like a good decision.
Parent - - By Uly (Gold) Date 2008-08-10 19:11
Father drew Rybka 3 on his 37th try but wasn't able to repeat a drawn game after hundreds of tries.
Parent - - By lkaufman (*****) Date 2008-08-10 20:09
Great! Now let's see what happens when Hikaru ("smallville") tries to do the same. By the way, has anyone even claimed to have won a game from Rybka 3 yet without computer assistance or takebacks or handicap?
Parent - By Uly (Gold) Date 2008-08-10 20:24

> has anyone even claimed to have won a game from Rybka 3 yet without computer assistance or takebacks or handicap?

Yes, due to bugs on the interfaces humans are able to beat Rybka on time. Of course this is beating the GUI so there hasn't been any claims of beating Rybka yet.
Parent - By Kapaun (****) Date 2008-08-12 21:55
I've had 3 games against Father, and he lost all three.
Parent - - By likesforests (**) Date 2008-08-10 05:20
Thanks for explaining the relevance of these events in relation to handicap games against Rybka. In the Aquarium handicap book you describe these handicaps like so:

Pawn and 1 Move = -325 (2863)
Pawn and 2 Moves = -450 (2738)
Pawn and 3 Moves = -550 (2638)

But the ratings corresponding to these titles is closer to:

GM =~ 2550
IM =~  2450
FM =~ 2350

Do you have any plans to update the numbers in your handicap book with this new data... or do you consider your original estimates more reliable?
Parent - - By lkaufman (*****) Date 2008-08-10 05:25
Those numbers were based on blitz type testing, and therefore are really appropriate for blitz games. I didn't write the Aquarium literature you cite, if I did I would have said this. On the other hand, Rybka actually scored 75% (about +200 Elo) in the games against those players at those handicaps, so statistically we might say that you need to be 2500, 2600, or 2700 to be evenly matched with Rybka at the three handicaps at intermediate levels like these matches (30'+30" or so).
Parent - By likesforests (**) Date 2008-08-10 05:29
That makes sense... thank you.
Parent - - By oudheusa (*****) Date 2008-08-10 20:36

As noted before, and with all compliments to your efforts so far, I have really grown tired of these handicap matches. It is clear to everybody that Rybka is a monster and that no human player is up to it. What are you trying to prove with these handicap matches? Even the upcoming match against GM Milov will be very one sided and you know it.

I would really welcome a centauer match at classic time control against a strong GM as suggested earlier. GM gets to use openings book, tablebases and a light engine like Fritz 6 for checking of tactical variations.

Any chance you may throw in two of those games in the upcoming match with Milov?
Parent - - By lkaufman (*****) Date 2008-08-10 20:48
     What I hope to prove from these matches is that the 3200+ ratings are not totally unjustified. They may be a bit inflated, but not substantially. I don't think that the Milov match will be one-sided, in fact I think that he is the favorite in the six handicap games and it's only the two normal games that may save us from defeat. Anyway there is no other practical way (short of million dollar sponsorship) for proving the point.
     As for a centaur match, it doesn't interest me very much, as it's basically a test of the GMs centaur skills, not of Rybka. But I'll be glad to run such a match or include two games of it in the Milov match if you or anyone else makes enough contribution to the prize fund to temp him into spending another day on the match.
Parent - - By oudheusa (*****) Date 2008-08-10 21:02
I am not sure who you are trying to convince of Rybka┬┤s rating of at least 3100. I believe you and I think most people in this forum. What is the current price fund? I am happy to contribute for two additional centauer games at classic time control if you promise that (provided they are interesting) you will organize the next match as a centauer match ))
Parent - - By lkaufman (*****) Date 2008-08-11 02:43
     There are still many people (perhaps not those in this forum mostly) who believe that Kasparov or Anand or Kramnik or Carlsen might still have a chance in a match with the best program, at least with really long time limits. So they probably think that the "real" rating of the best program is only 2850 or 2900. I would like to make it clear to everyone that this is nonsense. The prize fund is $3,000 for a lost match, $3,750 for a drawn match, and $4,500 for a won match, plus a paid simul. Let me know (private message if you prefer) how much you would be willing to contribute for the two extra centaur games, I'll relay the proposal if it sounds like a reasonable amount. As for a future match, I can organize a centaur match, but if you want a Grandmaster for it someone must be willing to sponsor it. Probably the most likely GM choice for such a match would be Dzindzi, because he is often in Washington anyway and has some experience with freestyle (with me). 
Parent - - By oudheusa (*****) Date 2008-08-12 14:59
Well then the purpose of the match is publicity, my objective was interesting chess ;)

The best publicity is a match against the winner of Anand / Kramnik. Now that Chessbase is supporting Rybka there is actually a chance you might pull this of with their help. However, I do not think either of them will agree to any type of handicap match, but a centauer match might interest them from a pure chess perspective. They played two centauer games against each other last year in Germany so I think they are interested in this type of play.

A nice option would be for them to choose their own engine, provided that is does not have a higher rating then they have themselves. Rating of engines to be measured by the CETG normal time control list.

This type of match could interest big name sponsers that have supported matches like this in the past.
Parent - By lkaufman (*****) Date 2008-08-12 15:19
Such matches serve several purposes at once. One is publicity, one is "interesting chess", one is learning things that might improve Rybka, and one (of particular interest to me) is to better estimate the "true" rating of the program on the human scale. Handicap matches are not an ideal way to estimate ratings, but they are better than just seeing how often a strong GM can manage to make a draw, which would mostly be a function of the opening book used. If you just want to watch really high quality chess involving humans and engines, just play over freestyle games. Centaur vs. engine is somewhat rare and might be useful for all of the first three goals, but is useless for the fourth one. But as I said I'll be glad to organize and run such games if we have a sponsor.
Parent - - By karmazen_robert (**) Date 2008-08-10 21:16

it could be interesting to publish the books in function of the used hadicap...  
I am interesting in the book used against Roman... without the f7 pawn...

those books are necessary so that rybka doesn't fall in an opening trap and the results of the match are not correct...

I don't know as me I will win to rybka, but neither you like it is possible to lose so clearly with those handicaps or odds...

Have you thought of publishing the books in function of the handicap in the web?
Parent - - By lkaufman (*****) Date 2008-08-11 02:46
The handicap books come with both the Convekta Aquarium GUI and the ChessBase Rybka GUI. However I update them for each match, so just as with any opening book you buy it will never be as current or complete as my own private book. As we had never held a match at Exchanged odds, I did not include them in those books, but I will of course make a book for this before the Milov match.
Parent - By karmazen_robert (**) Date 2008-08-12 21:42
OK. thanks for you answerd... and... I say that you do a good job with all books...

bye. from spain. Oliver
Up Topic Rybka Support & Discussion / Rybka Discussion / The Trilogy is complete!

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill