Junior-Fritz Match

Damn, I want to see Rybka demolish these pretenders. Junior won the World Championship at the Olympiad only because Rybka 2 was in beta at the time and needed more tuning and debugging.

Now Rybka 2 simply dominates the chess computer rating lists, at all time controls. I saw a rating list a short time ago in which the top 12(!) performers were variants of Rybka.

From that rating list that Mig linked to, we can see the following ratings:

Rybka 3110
Fritz 2932
Junior 2925

So Rybka is rated 178(!) points higher than Fritz and 185(!) points higher than Junior. It's not even close.

The closest chess engine to Rybka on that list is Zap!Chess Zanzibar with a 3040 rating, but that is still a 70 point gap.

Yah, although that's a little unfair because if an engine has a big rating lead it's likely it's other recent versions will be similarly strong(er). Still, clearly Rybka is a different animal, as I wrote here earlier. But as the saying goes, when you win can be as important as winning. A swiss tournament in which one loss basically ruins your chances isn't the same as the thousands of games the engine lists are based on.

Anyway, new versions are always fun to watch, although a six-game match is a strange thing in computer chess. During the same six-day time span the afritzionados will play sixty games to be more "conclusive" and then someone else will say you need 600 games. Meh, I don't have nearly as much interest in comp-comp if only because so much of it is about the opening books.

Is it just me, or is this match a pointless exercise? It's just publicity for Chessbase, but is anyone really interested? Not me...

Rybka's rating is based on much less games played, and really has not proven itself in tournament or match play. All the hype about Rybka is... hype of course.

@Parsnips:

Huh? Rybka's rating is based on playing long matches (50 games) against each rival engine. Rybka typically demolishes the other engines by scores of like 35-15 over 50 games.

The rating lists are based on match play and it has as many games as the others. Unless you consider a six-game match in public more "proof" than twenty or two hundred games in private. And it's done quite well in tournaments.

You can try it yourself if you like. I have. There is no doubt that Rybka beats other engines consistently. I've run Nunn matches (no opening book) between all the top engines and Rybka has no peer in comp-comp play. It's positional sense is far superior to anything else around, though still somewhat primitive compared to a human master in that department.

That said, I still use Fritz most of the time for work because it's faster tactically and that's about all I use an engine for.

So basically the #1 player (Rubka) is not the world champion. Now why does that sound familiar??

Need a unification match between the Chessbase and Chess Assistant world champions..

@StringTheory:

But like Mig pointed out, Rybka did not participate in the last WC. A beta version of Rybka 2 named Rajilch (after the author) competed but still had some kinks in it. The debugged Rybka 2 is the one with the 3110 rating, 185 points over Junior.

Go to one of the chess engine testing site and download the Rybka matches for yourself. Rybka just completely annihilates the other programs. It's as if Rybka is a 2900 GM while the other engines are 2700 GMs.

"So basically the #1 player (Rubka) is not the world champion. Now why does that sound familiar??"

C'mon, give it a rest. No need to keep the candle burning for Fischer _that_ long.

Rybka plays chess clearly better than Fritz, Junior or any other engine. I have run many engine touraments with and without opening books and Rybka is always on top. However an engine match of only six games or a world chess computer championship with only 9 games is another story since only a limited number of games are played. In such circunstances having a strong opening book is as important as engine chess strength itself. Junior and Fritz I believe have the financial resources to hire GMs to do the fine tunning in their opening books whereas the same does not apply to Rybka. What an unfair world!

This is interesting: There is near unanimity that Rybka is the strongest chess engine, and by a substantial margin. Alas, Rybka lack sponsorship support. I suppose that somebody could organize a public championship, which would be a match tournament that is transmitted live online, and which involves Fritz, Junior, Rybka, and a strong 4th program. The question is whether ChessBase would object to having its programs participate.

Rajlich can probably find a way to minimize or nullify the advantage that Fritz or Junior have by utilizing GMs to prep the opening books.

Rybka has a 175+ gap over Fritz, and can probably afford to spot the other programs an edge in opening prep.

Kramnik did a poor job against Fritz. Right now, at least, Anand seems best suited to be the human representative--although it's been a while since he's played a big match vs. the computers.

Well, $100,000 is a lot of money (even after FIDE takes its cut!) However, it is a mere fraction of the amount of money that ChessBase was forking over in order to get Top human opponents to sacrifice themselves on the altar of PR.

As you noted, ChessBase owns both programs, as part of its "stable" of Chess engines. So, it will be a case of one hand paying the other (less that FIDE commission); this may not cost them much money at all.

We'll have to see what kind of a show ChessBase puts on. I'm expecting that their Chess server will have very good coverage of the event, indeed!

=========================================
It will be staged and sponsored by FIDE president Kirsan Ilyumzhinov with a prize fund of $100,000 and take place in Elista alongside the final phase of the candidates matches (June 5-12). A hundred grand?! Hmm, what was the prize fund of the candidates matches again?

This must be the biggest prize fund ever for a machine-machine competition and I'm more than a little surprised by it – unless it's largely funded by ChessBase, which would make sense as a promotion since they publish both programs.

As far as I'm aware, Rybka has participated in 7 official tournaments since it went commercial. Winning 6 of them, sharing 2nd place in one. Here they are, with the most well-known competitors mentioned:

1) 15th International Paderborn Computer Chess Championship - 5½/7, 1st place ahead of Gandalf, Zappa, Spike, Shredder, Fruit.

2) CCT 8 - 8/9, 1st place ahead of Zappa, JUNIOR, Hiarcs, Spike, Fruit, Diep.

3) 6th International CSVN Tournament - 8½/9, 1st place ahead of Shredder, Gandalf, Diep.

4) 14th World Computer Chess Championship - 8½/11, shared second with Shredder, behind JUNIOR (9/11), ahead of Zappa, Spike, Diep.

5) 26th Open Dutch Computer-Chess Championship - 9/9, 1st place ahead of Loop, Hiarcs, Fruit, Gandalf, Shredder.

6) 16th International Paderborn Computer Chess Championship - 6½/7, 1st place ahead of Shredder, Gandalf, Spike, Diep.

7) CCT 9 - 6/7, 1st place ahead of Hiarcs, Spike, Naum, Loop, Zap!Chess, JUNIOR, Diep.

Rybka lost two games in those 7 tournaments, 1 against spike in her first tournament, and then the one against Shredder costing her the World Championship.

Total score: 52/59.

Besides that, Rybka tops all relevant rating lists (based on hundreds and thousands of games) with an unprecedented margin.

The careful reader will have noticed one engine missing from all these tournament results: FRITZ.

But of course, when your flagship engine by all objective measurements are ranked along with a pack of engines rated 150 ELO behind the leading engine, it looks tempting to shield it from any potential embarrassments. Much better is to find a sponsor for a match against Kramnik, and now an exhibition match against another in-house engine :-)

What happened to Hydra?

it looks totaly ridiculosus match. There is no point in spending money and effort on it since chess needs them in other directions. I will not even look at their site.

Speaking of Hydra. Thats the match i would like to see HYDRA vs. Rybka...Sort of Like Ali Frazier...Sugar Ray Leonard vs Roberto Duran....

I've been using Rybka to analyze games and opening plans the past few weeks and found it prefers to wreck its own pawn structure for piece activity, often going into a bad endgame as a result. However, if I set the mode to "very pessimistic" it will play much more positionally sound moves right from the opening. It would be interesting to see what effect this would have on its play against other programs.

I think chessbase is probably planning to release Junior 11. "The fine tuned version that beat Fritz 10"..

Btw, Toga 1.3 (the free engine) seems stronger than Fritz too...

Though I agree that Rybka is stronger than the others, watching Rybka games is boring. When Junior plays, it's like in-form Morozevich!

My conspiracy theory is:
Chessbase is perhaps the most important provider of chess news, and Ilyumzhinov is funding the match and putting the name of FIDE behind it, so that (or at least in the hope) that Chessbase articles become less harsh to FIDE operations...

Mig's point that Fritz is still more useful for many purposes is good - I would guess the top players in their preparation have no particular reason to rather use an engine that is stronger positionally but still massively weaker than they are themselves. This is also why I am not so sure that Rybka would be a significantly tougher opponent for a top human player than Fritz is...but perhaps it is so.

This is a pointless match. 99% of users know that Rybka is the program to beat and no amount of hype from Chessbase is going to change that perception.

Of course, their goal is to generate some publicity, not to prove once and for all what is the strongest program. But for 100 000 they could have run a very decent tournament and generate far more publicity for it too.

I don't understand the points repeatedly made by people (including Mig) about chess engines not having a master's "positional understanding". It's just not true, but EVEN IF IT WERE TRUE, the logical conclusion is that we have GROSSLY OVERRATED the "positional" aspect of the game; that "positional understanding" is not very important in chess.

How else can anyone explain computer dominance in chess?

It's time to accept the fact that computers play better chess than humans. The results prove it: how many times does a player need to beat you in order for you to accept that, in spite of his apparently "weak" play, he is a better player than you are, given that he keeps winning consistently?

With all due respect, a recurring theme whenever Mig has mentioned Rybka is that, whatever Rybka has going for her, it does not (necessarily) translate into being useful for the user. Taken at face value, such a statement would to me be pure nonsense.

In my opinion, Rybka is the most valuable engine (compared to competitors) the world has seen for many years.

I am sure that practically all top GMs use Rybka as their main assisting analysis tool. I am sure that when Cheparinov sits up all night cooking up novelties for Topalov, he has Rybka running as the main engine (a funny store is Frederic Friedel fiercely arguing with Topalov about whether Rybka is stronger than Fritz). I am sure that all top corr. players use Rybka (or they would not be top players for long).

Regarding the claim that Fritz is "tactically faster", it is a truth with a zillion modifications. Almost by definition, Rybka is the fastest engine available, since it finds the stronger moves faster, on average. It's true that Fritz (and others) is faster at suggesting some wild attacking moves. But then Rybka will be the fastest engine to resolve the resulting positions (the idea that Rybka is a slow searcher is a myth stemming from her low reported node count). Besides, Fritz will far too often suggest moves or continuations that just don't work (because it has far too big king safety penalties), so you will actually end up wasting time by following Fritz suggestions first instead of Rybka suggestions, on average.

Rybka is simply the most reliable engine available. This also means that for weaker players seeking to learn what was really going on in their OTB games, Rybka is the best choice. You can rely on her evaluations 10 times as often as you can rely on Fritz.

Finally, if you are really looking for the tactically fastest engine, the engine of choice would be Rybka WinFinder, which is part of the Rybka package. This has been documented countless times by people measuring solve times in tactical test suites.

Enough fanboi talk for now, I just wanted to speak up my mind at least once here :-)

The chess world lacks corporate sponsorship, and the give the money to machines? And Fide approuves? For sure they know how to organize a world championship.

Wake up people! Rybka vs either one of those engines would be a laugher and not interesting at all. It's really better to select two engines rated close together if you really must hold the match at all. Of course it's only a publicity stunt for the benefit of Chessbase.
Jim

Why a match between two chess programs is really interesting ?

"Why a match between two chess programs is really interesting ?"

Because there might be some nice chess played? Admittedly it seems a lot less interesting than a human vs human duel, though I'm not quite sure why. If computers place more emphasis on tactics shouldn't this lead to a more exciting game on average? It's gotta be some sort of craving for the drama, confrontation, or psychological component that are absent when machines are involved.

Given only the move sequences from start to finish, i.e. without any prior knowledge or the help of a program, I wonder how many of us would be able to reliably determine whether the participants of a game are man or machine? Might be a half-interesting experiment if you could find enough games that are high quality but relatively unknown.

I hope Chessbase is sponsoring the match; if not, what a waste of money by Ilyumzhinov. Chess without the human personality behind it is rather boring; I'd rather watch 10 Kasparov-Kramnik blitz games than 100 computer-computer G/75 games.

But seriously, Kasparov still plays blitz, right? For a winner-take-all $100,000 prize, I think both parties would play. :)

@Cynical Gripe

The moves of the computers are just an output of optimized algorithms written by humans.
If you change a single chess rule we can easily play chess with that rule changed : however, you should rewrite all the s/w in order to have the computer match played !

You can easily find out whether a given game is played by computers : they cannot use intuition and mature kind of thinking as humans. To replace them, they use databases made by humans!

I think all those things are not good signs for chess sport.

Btw does anyone know what hardware are they playing on? It's the same for both sides yes?

george,

I'm not so sure it would be easier for humans to adapt to rule changes. The software would have to be modified, yes, but it's quite outrageous to suggest a complete rewrite would be necessary. To test this theory couldn't a top GM play a top engine (or one with similar ELO) in a Fischer random match?

Unless your ELO is above, say, 2200 or so, I highly doubt you could "easily find out" just by replaying the game. And why can't computers use intuitive and mature thinking? They are programmed by humans as you mentioned. Just because most programs follow a certain mold doesn't mean it has to be so, but why fix it if it ain't broke? I'd wager a good amount that engine programmers could, and likely have, attempted to incorporate things like riskier sacrifices into their game, but it just didn't work out in terms of wins and losses, probably because other engines don't get rattled by that sort of play.

The list mig pointed to also shows the direct scores:

Rybka - Deep Junior 10 26-5 (+23 =6 -2)
Rybka - Deep Fritz 10 24.4-5.5 (+19 =11 -0)

Fide calls the forthcoming match "the Ultimate Computer Chess Challenge". :-)

While some of the comments here have been rather negative, the computer match actually promises to be quite interesting. For those of us who began by playing chess but later went into computer-oriented fields, it will provide some fascinating insights into the relative abilities of different types of algorithms.

On a related topic, most players here are probably familiar with IM Danny Kopec, who is a professor of computer science in New York. He has written some brilliant, but very readable papers on search algorithms. If you would enjoy reading some of them, his website includes a list of publications, many of which can be downloaded. The url is:

www.sci.brooklyn.edu/~kopec/

@Dave50, six games (only) in a match of that type will provide insight into opening preparation, more than into algorithm comparison. I think openings & statistical luck will decide this short match between two engines of not identical, but similar strenghts (the different style is undisputed).

For rating lists like CCRL, engines are compared with identical opening databases, which do not contain engine-specifici tuning, and on the same hardware. These conditions are realistic compared to the normal chessplayer's practical use of engines. Anybody who really wants to study how Fritz 10's engine algos work against Junior 10's algos, will prefer the games to download from there:

http://computerchess.org.uk/ccrl/4040/games.html#by_engine

Both Deep Junior 10 and -10.1, on 4 CPUs, have played 30 games each against Deep Fritz 10 on 4 CPUs, up to now. That means a total of 60 games for that comparison. The different time control 40/40m does not make a significant difference to 75m+5s in terms of win/loss chances, neither hardware differences within relatively small margins of different speed.

It can be an exhibition match and a kind of showcase event for computer chess, but both Rybka (and most probably some more engines) and Hydra are stronger.

During the last decade, Shredder has won about as many World Champion titles in computer chess as Junior has (it also ranks higher than Junior now, on most engine rating lists). But if it's about such "big" matches, it were always Fritz vs. Junior, Fritz vs. Kramnik, Junior vs. Kasparov, Kasparov vs. Fritz, Kramnik vs. Fritz and again Fritz vs. Junior. Where was Shredder, where was Hydra (except vs. Adams), where is Rybka, Hiarcs, etc.? Yes there were events where they participated, but never in such big ones and often without ICGA and/or FIDE support.

Are there people who accept only the same two engines in the whole world, again and again, and ignore all the rest, no matter how the true strength relations are? - A few people have decided who to get a money shower of 40 to 60,000 $

It stinks.

Permanent Brain,
Though I agree with most of your arguments, except the effectiveness of CCRL-and-alike testing.
What is the purpose of testing Junior with the same book as Fritz? Who says that no engine-specific tuning book does not have very, very large impact on playing strength? Would Fischer have the same win/loss ratio if he played Petrosian's openings? Would Kasparov if he played Karpov's?
I have looked up some matches of Junior vs other engines in those testing sites. Half of the games Junior starts with 1.c4 (or one-third), and gets an absolutely lost poition right after opening. On the other hand, when it is in "its own" positions it is extremely strong.
My point is that for a very specific engine like Junior, testing it without a very fine-tuned book is almost worthless. That's why they never have lead the ranking lists, but have won the WCH more than others.

@PlayJunior, you may be right, to some extent, about positive effects tuned books can have. But if a chess program is provided as a training and analysis tool for a chessplayer, it is HIS repertoire and the openings HE wants to practise, which are important, not those the engine is especially good in. That is why engine testing with generic untuned books - preferably with a wide repertoire - is much more realistic relative to a normal chessplayer's practical use of the program. The "engine sport" where engines & opening books fight against each other, is something different but a freak's niche, only.

To be more on topic, we can also consider that a "book win" in the upcoming match wouldn't really tell us much (or almost nothing) about the strength relation of Fritz vs. Junior, in such a particular game. I always disliked it very much, incl. world championships, if book cooks decided games. Such games aren't really computer chess competition.

I think, an engine which does NOT depend on a very specific extra-tuned book to be successfull and can support the user in many openings, is much better than another engine which is limited to a very specific repertoire to be competitive.

Btw. just today, I read an info about the Junior 10 book which may be interesting for you (but it's in german language):
http://f50.parsimony.net/forum202220/messages/23532.htm
It seems that after 1.d4 Nf6 2.c4, the J10 book (as Black) has a variation which does not provide any variety up to the 15th move (always only one 100% black reply). That makes it vulnerable for computer opponents which could try to repeat a win against this line. (I couldn'T crosscheck this info as I don't have that book and I'm only quoting it.)

Sorry if these details are considered a bit off topic, but it fits to the "opening preparation" aspect of big computer matches, too.

Permanent Brain,
I agree with you on some points, i.e. that chess engine's main aim is to aid analysis. Probably, from this point of view general-book testing is reasonable. On the other hand no one uses Junior for analysing rook endgames. Everyone knows that it can lose any, even the most drawish endgames. You use it for specific positions, and when you want to know how strong is it, you would like to know how strong it is in THOSE positions. So, one can argue that those general-purpose testings favour general-purpose engines, and specific ones suffer because of this!
My opinion is that Junior is extremely strong, and is going to dominate Fritz.
By the way, I've always been asking myself-why doesn't Junior team write a "separate" endgame engine and "switch" to it at endgame? This could add them some 100 points.

"What is the purpose of testing Junior with the same book as Fritz?"

We at CCRL test engines with a neutral genuine book, because we aim to test playing strength of the engines and only that. So we decided to minimize book inteferences by giving all engines a short (12 moves max) book.

Andreas, I hope you understand what is my point. Bobby Fischer was playing only 1.e2-e4, and if you forced him to play 1/3 of his games with 1.d4 and 1/3 with 1.c4, (arguing that is a "genuine neutral" repertoire), would his results be the same? Why?

Junior has won so many WChs, while it rarely was on top of any rating list(maybe sometime, some ooold version was on top on SSDF, I dunno). I think this supports my point of view.

Fritz is a stronger program clearly.

Another program that seems to be very aggressive and likes the attack at ANY cost is Naum 2.1. Out of all 50 or so of my engines it seems to be the one who most palys like Tal.

Another program that seems to be very aggressive and likes the attack at ANY cost is Naum 2.1. Out of all 50 or so of my engines it seems to be the one who most palys like Tal.

Something interesting to take into account when talking about pure engine "strength"

I'm running a Rybka v 2.2n2 on Fritz 11, each using the same opening book which I researched online using Openingbook ratings

"CompMaster Beta1.0.ctg"

Fritz 11 vs Rybka after a few tests such as 10/10 1 min games rates itself as 32 points higher than that particular Rybka version. It's Fritz 5.5/10 and Rybka 4.5/10,

And then the exact same result with 10/10 15 minute games, fritz being slightly better.

HOWEVER, when running engine matchs using opening databases, Nunn, under blitz time controls of 5 minutes per engine, Fritz only won ONE of the entire set-ups, with rybka winning something like 15.

This tells me that the actual rybka engine is far superior at calculating things, and quicker especially on blitz games.

At the moment I'm running 50 games using the Nunn2 opening database of Fritz 11 vs Rybka with 40/4 time settings to make up for the fact that Fritz is HORRIBLE on the blitz games vs rybka or at least appeared to be to me, and also would be at 1 minute left on time out of thier 5 minutes with Rybka still at 4 minutes 20 seconds left out of 5 minutes.... that's kind of an important advantage towards Rybka in my opinion because the fritz engine has to rush the rest of it's own moves compared to what it would like to for most of every Nunn game.

The biggest question I have however is, if rybka does so well in the the opening database games where neither game uses an opening book because the starting positions are set up, then why is it doing worse than Fritz in Engine matchs after they have both used thier last opening book move, even at 1 minute timed games when it seemed to me that Rybka has an advantage for making quick moves.

And the second question I have is about the opening book CompMaster Beta1.0.ctg. Should I be having all my chess engines use this book? Does that make engine vs engine games more fair?

And the third question is about the opening book option in the Fritz 11 platform about it's learning option for opening books where it doesn't favor positions where it has lost from before. I have no idea how this works, if it is program specific, if the adjustments made for one program will change the .ctg file for all programs, and if un-loading the opening book for an engine that has been "learning" from CompMaster Beta1.0.ctg, and loading up it's own engine book will have said program effectively forget the changes it had made for itself using CompMaster Beta1.0.ctg and force it to re-learn everything it had already learnt the next time I load this book back into the engine.