Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Come on Vas! Win the next 5......
- - By jps7 (**) [in] Date 2007-09-24 09:12 Edited 2007-09-24 09:16
Hi,

I still don't believe that Zappa can do it. But if Zappa can win 3 in a row, why not rybka the next 5. Just play on your strengths. As black, clearly Zappa's strategy is to stretch the game longer and longer and finally win the endgame (see 180 moves game 4). Zappa is clearly taking advantage of the deeper search on 8 processors due to better scaling. I think that knowledge based engines will have problem in scaling on 8 processors as compared to heuristic search based engines due to knowledge representation problems. It seems that Zappa team has studied most of the games where Rybka has lost (may include Freestyle Rybka-Rybka games as well) and trained Zappa Mexico on most of the positions which Rybka somehow does not like. I also feel that shredder's tactical strengths have also been included in the Zappa and new Zappa Mexico is playing positionally and tactically both with better kingside attacks. May be the shredder bit bases are also helping Zappa in the end game accuracy as I have seen Deep Shredder 10.1 playing endgames quite well. So, the only option available with Rybka is to better its search and improve the endgame evaluations. If the knowledge base can also be increased (may be with Capablanka games) that can well help but I dunno how? Vas has done well with French Defence in Freestyle games. So why not give it a try here too. Good Luck to the Rybka team (off course to the Zappa team as well). I surely know, at least Rybka will level it 5-5.
Parent - - By Uri Blass (*****) [il] Date 2007-09-25 04:46
I do not agree with your opinion.
I think that zappa is the knowledge based engine and not rybka.

My opinion is that the relative advantage of rybka with less processors is superior search when with more processors zappa can close most of the search gap and take advantage of it's superior evaluation.

Uri
Parent - By dareapa (**) [us] Date 2007-09-25 05:02
Was Zappa Mexico created for the match?  Or was there always gonna be "Zappa Mexico"?
Parent - - By JhorAVi (***) [ph] Date 2007-09-25 06:26
Zappa has roughly twice more KNPS than Rybka. Isn't the slower one supposed to be the knowledge based engine?
Parent - - By M ANSARI (*****) [kw] Date 2007-09-25 06:53
I did a 100 game 5 0 gauntlet tourney with Zappa Mexico X64 against Rybka 2.3.2s and Zappa only managed a 16% score while Rybka scored an impressive 84%.  This was on a vapor cooled 4.3 Ghz system ... I really don't understand what is going on in Mexico ... could be that the version being sold is not the same as playing ... or that the engine Rybka is playing is not properly tested ... or simply that Zappa scales tremendously on 8 cores ... or that Zappa simply has gone on a good run.  In my tourney Zappa was not able to win a single game.
Parent - - By Intagrand (*) [gb] Date 2007-09-25 07:01
Repeat the 100 game match at the time control of 60+20 or 120+40 and things will be completely different. It's a fact that Zap plays much better when it has plenty of time to think. In 5 0 games, Rybka searches far deeper than Zap and Zap loses due to being out searched - things beyond it's search horizon. When Zap has enough time not to miss tactical shots, it starts to play on more equal terms with Rybka.

What's going on in Mexico is that Zap and Rybka are playing 60 + 20 on 8-way computers and not 5 0 utilising half the processing power of a quad core machine.
Parent - - By M ANSARI (*****) [kw] Date 2007-09-25 08:19
You could be right ... but really I think that with chess engines usually a 5 0 match gives a very accurate estimate to the superiority of one engine to another.  It could be that Zappa is an exception to something that I have noticed in more than 10 years of testing engines ... but I really doubt it.  What is considered 5 0 time control on a 4.3Ghz Quad today would probably have been equivalent to a 100 0 time control a year or two back.  Since in my tourney they are playing identical hardware and same book then I would guess something else is in play.  After watching Erdo book draw yesterday, it could be that Zappa simply has a large gain in ELO due to Erdo's fantastic preparation.  My guess is that it is a combination of Rybka not having a properly tested engine and Zappa scaling much better than Rybka on 8 cores (Rybka seems to not suffer from MP efficiency with 4 cores) and good preparation by Erdo.  Beta engines need to be tested thoroughly otherwise improvements can backfire.  I remember one version of Rybka I tested at one time was 300 ELO points weaker due to some very minor bugs.  It could also be that Zappa simply have a different version of Zappa Mexico running ... but I really doubt that.
Parent - By Intagrand (*) [gb] Date 2007-09-25 08:34
One thing you could do is to compare the difference in search depths between Rybka and Zap at 5 0 games and then again at 60 20. I'm sure Zap is being completely out searched in 5 0 games which is why any engine would lose. It's simply not searching deep enough to compete with Rybka at that TC. Please do try a long test and you might be surprised that Zap is one of these engines that plays better with more time. If Rybka beats Zap with a score of 84% in 100 games at 60+20 on your machine, I'd be very surprised.
Parent - - By turbojuice1122 (Gold) [us] Date 2007-09-25 11:14
Yes, I think it's been known for over two years that Zappa is one of these rare exceptions.  In 60'+20" games on my machine, Rybka is struggling to equalize the match.  Of course, one very major key is that I'm having them use books that are suitable for them with book learning ON--I don't see the sense of forcing the engines to play openings that they play weakly, thus multiplying the random statistical anomalies by large factors and not giving an indication of how these engines play in match situations.  While Zappa scales well with more cores and 64-bit OS, it also performs extremely well on 1 CPU relative to other engines if you just give it long enough to think.  Rybka, on the other hand, has such a great evaluation at one ply that it uses this to great advantage in blitz games, but the relative advantage to other engines will be less in match conditions at tournament time control games.
Parent - By Uri Blass (*****) [il] Date 2007-09-25 12:02
I do not believe the story about great evaluation.
I think that it may be the great search of rybka that is winning at blitz.

Programs get many plies at blitz today and not one ply so you cannot use blitz results to compare evaluations.

Uri
Parent - By Lukas Cimiotti (Bronze) [de] Date 2007-09-25 08:42
I played 6 long games against Erdo before the event - i won only one game, the rest were draws.
We both played on 8 cores, he used Zappa Mexico on an 8x2.66 GHz Intel, i used Rybka 2.3.2j4 (basically quite equal to the version, that played the last games)on an 8x3 GHz Intel. So i knew, it would become tough.
Parent - By er (**) Date 2007-09-25 12:23
Maybe it is Erdogan's fine book
that makes the difference in Mexico.

ER
Parent - By Permanent Brain (*****) Date 2007-09-25 07:08
As rumour has it, Rybka doesn't display it's true internal node number like (presumably) other engines do. It may be much higher. Node rates tell nothing about engine quality anyway. Also, I think that the old simplification "small node rate = knowledge based, big node rate = search based" is wrong, or at least not 100% true. Programmers could explain that better.

I do not doubt that Rybka (even 2.3.2a) maintains number one, but a match is a snapshot in time over a small number of games, only. Also, Zappa being number two on 2 cores or more for some time now, deserves increased popularity. Computer chess in general needs much more public relation. Related to this match, which attracts more "general" chess players as it seems, I notice that some don't even know Zappa the World Champion of 2005, or for example also never heard of Fruit/Toga. That needs to be improved.
Parent - - By Uri Blass (*****) [il] Date 2007-09-25 11:53
No

NPs mean nothing because it is easy to change them by printing different numbers.

changing one line in zappa code from something similiar to printf("%d",nodes) to something similiar to printf("%d",nodes/10) is clearly enough.

Uri
Parent - - By JhorAVi (***) [ph] Date 2007-09-25 12:25
May I know one logical reason why Vas should misinform us with a wrong KNPS?
Parent - By Vempele (Silver) [fi] Date 2007-09-25 12:30

> May I know one logical reason why Vas should misinform us with a wrong KNPS?


Maybe because he has already admitted he does so. The conventional definition of nps = nodes (searched) per second. Vas says Rybka doesn't count all nodes.
Parent - - By jps7 (**) [in] Date 2007-09-25 15:16
Hi,
Somehow, I read somewhere that Rybka is the first Knowledge based engine. However, I am still not clear how a knowledge based engine differs from heuristic search based engines or a brute force engine. Also, how a heuristic search engine differs from brute force engine or are they same.
Parent - - By The Wizard (***) Date 2007-09-25 15:20
Hi,
    Zap won the 2007 KOE http://www.kingofengines.co.uk/index1.html which is 180 minutes for the game and not Blitz ... It was not even the Mexico version

Parent - - By turbojuice1122 (Gold) [us] Date 2007-09-25 15:25
Yes, but the version that won was running on one of Suj's monster machines; I forget whether he used a 16-core machine or 32-core machine for this tournament.  Rybka was running on a quad.
Parent - - By Arrière Pensée (Gold) Date 2007-09-25 16:16
Hi turbojuice1122,

This game Rybka is being great deal more guarded in her moves.
Parent - By Arrière Pensée (Gold) Date 2007-09-25 16:39 Edited 2007-09-25 16:47
"SHAMANISM
WORKING WITH ANIMAL SPIRITS

We are many.....and will take a moment.....                  
          ..... our wisdom will appear to you before we do."

Allow little Rybka to take to form of the Whale in this game against her foe Zappa Mexico

"Whale's Wisdom Includes:

Record keeper for all eternity
All knowledge associated with voice
Psychic and telepathic abilities
All aspects of the sea
Beauty of movement "

Let Rybka Prevail
Up Topic Rybka Support & Discussion / Rybka Discussion / Come on Vas! Win the next 5......

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill