Not logged inRybka Chess Community Forum
- - By Jack-Ireland (**) Date 2008-01-03 03:24
Hi All,

To begin with, Happy New Year to all.
Did anyone read the latest workshop on chessbase? Lopez compares Junior to Fritz, Hiarcs and Shredder.
Junior is the most tactical one and Shredder is the most positional one. I would be interested to read how rybka compares to these - not in playing strength, but in playing style.

Best, Jack
Parent - - By Uly (Gold) Date 2008-01-03 04:06 Edited 2008-01-03 04:10
I've said this over and over, but I'll say it again. To me, Rybka's style is "Hyper-Solid". On a given position, she uses the first plies to see what is the move that more likely will be more solid, and once she knows this, she expends most of the analysis time checking the solidness of this move, so the chance of speculation is greatly reduced.

Rybka throws the whole house at this move, and only if this move proves worthy of "refuting" anything, she plays it. And only if this move doesn't look as good, she'll consider something else.

You can check this by comparing the time Rybka gives to the main move and to the rest of the moves, and compare it by the time other engines give to the main move and to the rest of the moves. The relative time that Rybka gives to the one considered best move is overwhelming, and she barely spends time on the rest of the moves. This gives Rybka plenty of extra time to examine the main variation very deeply, and it out-searches most other engines (due to Rybka's very good evaluation.)

This produces a playing style as solid as it can get, with the hopes that the opponent (that is not solid) will make inaccuracies on its moves. It happens, and once Rybka is on a winning position, she knows what to do, to the ultimate consequences and makes the opponent lose in the end.

Also, when for some reason this technique doesn't work, and Rybka ends in a losing position, this losing position will still be solid, so Rybka is very good at defending it.

I think that this is the reason that Rybka is at the top, though I think that Vas stated that he will make Rybka 3's playing style differently, more aggressive, I hope.

EDIT - Other users have agreed with this, adding that now, Shredder 11 and Fritz 11's styles are more solid, more "Rybka-like."
Parent - - By 8lrr8 (***) Date 2008-01-03 04:26
to me, "solid" is playing moves that are a tad on the passive side.  i think of kramnik or leko.  "solid" moves tend to up your draw percentage.  rybka doesnt play such moves.

"optimal" would be a much better adjective, ;-)
Parent - - By Uly (Gold) Date 2008-01-03 06:47

> to me, "solid" is playing moves that are a tad on the passive side.


To me, Rybka seems just like that before the opponent's inaccuracies, I often am bored seeing Rybka play. Rybka just minimizes her chances of blundering while maximizing the opponent's chances, but does not attack the opponent directly until then.

I would put her on the other side of aggressiveness. Of course, her style changes when she's after something, but I'd like to see this style from the beginning. Rybka is just too subtle to me.
Parent - - By 8lrr8 (***) Date 2008-01-03 06:51
"...but I'd like to see this style from the beginning."

it may very well be that u may never see this if such a "style" proves to be sub-optimal.  as vas once said, u cant just attack out of nowhere (lest u get your ass handed to u on a silver platter).  u need to first develop your pieces before u can (successfully) launch your assault.

doing so prematurely is a sure way to defeat.
Parent - - By Uly (Gold) Date 2008-01-03 20:25

> as vas once said, u cant just attack out of nowhere (lest u get your ass handed to u on a silver platter).


As we're talking about style, I meant that when most moves draw, Rybka can pick the most aggressive of them. When there's such a possibility, my impression is that Rybka sees that with perfect defense, the attack does nothing, so she just doesn't attack. However, she's not playing against a player that will play the perfect defense, so in my opinion, she should try it anyway.
Parent - By 8lrr8 (***) Date 2008-01-04 01:06
there may be nothing wrong going on at all.  it may just be that rybka doesnt like the end position of the tactical combination, but rather prefers its current position's eval.

to to be victorious, u must first learn how not to lose.  learn first to become undefeatable.  only when u have achieved that should u focus on how to win.
Parent - - By onursurme (***) Date 2008-01-03 11:25
of course, this needs an ability to quickly find the best move in the first plies of search :) , only after this, Rybka can test that move at the rest of her thinking time.
you can see it by selecting the level of 1 ply depth for Rybka and play againt her, you'll see she is still good at 1 ply depth.
Parent - - By Uly (Gold) Date 2008-01-03 20:28

> of course, this needs an ability to quickly find the best move in the first plies of search


But there are many "best" moves in a position.

Also, I've heard that Rybka does some extensions as to make Ply 1 be basically Ply 4 of other engines.
Parent - - By richbell (**) Date 2008-01-03 21:19
Wow, Thanks for the insights Vytron. It was interesting to know Rybka's style in from your simple explanation.

It will be great if Vas add to this his view on the secret of Rybka's strength and how to plans to improve it in 3.0.

I saw Alex, Naumov mentioning in talkchess he almost could achieve a version to beat Rybka, but it keeps loosing with other engines. So what do you think his idea is? :-)
Parent - - By Uly (Gold) Date 2008-01-03 21:24

> So what do you think his idea is?


The weaknesses of Rybka have been discussed at another thread, so perhaps his plan is to  get rid of these weaknesses.

Without the weaknesses, Rybka becomes stronger against everything.
Parent - - By richbell (**) Date 2008-01-03 21:29
I think i didn't convey it correctly, He says the version that beats Rybka keeps loosing to other engines. What kind of strength that kinda engine will hold then?
Parent - - By Uly (Gold) Date 2008-01-03 22:31
Zappa Mexico beat Rybka in a match of 10 games with the philosophy: "If you beat Rybka, you beat everything else", and that probably remains true (At least, with a big hardware and a superb book).

The only personality I can conceive that would beat Rybka, is one very good at setting up a hidden King attack that Rybka doesn't see until it's too late. This engine would need also to not make inaccuracies while doing so, as weakening its position will cause Rybka to attack and then this engine spends the time defending against Rybka's attacks and doesn't have the time to make the King attack.

But, an engine solid enough to keep Rybka on the solid state, that is also able to hide King Attacks that not even Rybka can see, would probably be good enough against all other engines, too.

So, I don't know what Alex Naumov's idea would be, for an engine that is only good against Rybka.
Parent - - By richbell (**) Date 2008-01-03 23:32
Vytron, what you say sounds right. I am little intrigued with what he said and how to achieve that kind of Anti-Rybka engine, which still plays poorly with other engines. :-).
Parent - - By Uly (Gold) Date 2008-01-04 00:21
My opinion is that this engine should have an "Anti-Rybka" option:

[  ] Anti-Rybka

You only switch it on when playing Rybka and keep it off for the rest of the engines.
Parent - By richbell (**) Date 2008-01-04 01:02
I am not sure if you saw the actual post :-)

This is what was actually suggested to Naumov.
Parent - - By ozziejoe (**) Date 2008-01-09 08:53
but rybka's ply =1 is really ply =3.
Parent - By Uly (Gold) Date 2008-01-09 16:24

> but rybka's ply =1 is really ply =3.


I think Rybka "ply=1 really ply=4" (1+3) was confirmed due to Strelka (Rybka's clone without the obfuscation.) But it does not prove that future Rybkas (beyond 1.0 BETA) do the same (So we don't really know.)
Parent - By Carl Bicknell (*****) Date 2008-01-03 11:40
BTW - I question the accuracy of the review you picked from chessbase. He notes hiarcs as 'positional' and although this is a vauge term the prog is surely more aggressive than Fritz (though I trust Fritz much more). He also notes Shredder as "Petrosian - like" which used to be true but V10 seemed pretty aggressive to me.
Parent - - By 8lrr8 (***) Date 2008-01-04 01:17
the concept of style is a human construct.  quantitatively speaking, the spectrum runs from suboptimal to optimal.  optimal here means given a won position, the winning side mates in the fewest # of moves possible.  in the event 2 or more lines mate in the same number of moves, u can simply flip a coin (or roll a dice) to choose which line to play.  the corollary is the losing side prevents mate in as many moves as possible.  in a drawn position, the side w/ an advantage chooses the line that has the narrowest draw window.  again, if multiple lines are equally narrow in draw probability, flip a coin.  subsequently, the side w/ the disadvantage also does the same (i.e. it chooses moves that against a suboptimal player, gives the most likely chance for said player to blunder).

talks about "aggressiveness," "human-like playing style," "positional," "tactical," et al, are nothing more than homo sapien sapiens' faulty interpretation of game-theoretic objectivity.
Parent - - By Uly (Gold) Date 2008-01-04 01:30

> talks about "aggressiveness," "human-like playing style," "positional," "tactical," et al, are nothing more than homo sapien sapiens' faulty interpretation of game-theoretic objectivity.


Probably, but this also applies to humans. So I can say that one GM plays more aggressive than another, and I don't see why can't I say the same for engines.
Parent - - By 8lrr8 (***) Date 2008-01-04 01:44
but all "aggression" means is tendency to capture or threaten to capture, or tendency to check or threaten to put in check, right?

if so, this tells us nothing about how strong the GM/engine is.  which means it's a "useless" metric.  :-)
Parent - - By Uly (Gold) Date 2008-01-04 20:18

> which means it's a "useless" metric.


But it is useful. When an aggressive player makes threats against a weaker opponent, the weaker opponent isn't going to be able to stop all of them, so the aggressive player is going to reach a winning position earlier. When the stronger player is "Super-Passive", it's not going to beat anyone because he will not attack even if there's a weakness of the opponent. So style has an effect on the game.

I think that Rybka would perform better if she was more aggressive, just because most of her opponents are weaker, and are not going to find the perfect defense that she knows that exists.
Parent - - By Kapaun (****) Date 2008-01-05 10:00
I think that Rybka would perform better if she was more aggressive, just because most of her opponents are weaker, and are not going to find the perfect defense that she knows that exists.

Actually I doubt that. At least not the Rybka version we all know at the moment. If you tell her to play more aggressive (by parameters), she seems to play weaker all in all...
Parent - By Uly (Gold) Date 2008-01-05 23:04
But it's because it's playing against another Rybkas, so being less solid just harms. I propose a "Solid-Aggressive" approach, in where you attack the opponent while getting sure that you don't open holes in your position, yourself. Maybe this is not possible, though (I don't have the 8 cores that Zappa Mexico needs to check this.)
Parent - By Hetman (*****) Date 2008-01-05 19:56
'Supper-Solid' ? player will win in endgame. :-). There is the time to exploit weaknesess of the position without big risk.
Look at Karpov, T.Petrosian games. They were not aggressive players but were WCCH. I do not know what to think about Drawnik games. ;-)
Rgds
Hetman
Parent - By 8lrr8 (***) Date 2008-01-09 03:01
1. i think u are referring to contempt.  and yes, rybka currently has suboptimal contempt.  this is a difficult thing to implement, i believe.

2. there is no such thing as a "super strong, super passive" chess engine.  so u cant make that comparison.
Parent - By Vasik Rajlich (Silver) Date 2008-01-04 10:42
Human-like playing style can be defined and measured without almost any controversy. It's a simple matter of correlation between human moves and engine preferences. It would be interesting to do it and I'm surprised that it hasn't been done yet.

Things like "aggressiveness" could also be defined and measured, although the definition wouldn't be completely routine.

Vas
Parent - - By boo! (**) Date 2008-01-04 20:54 Edited 2008-01-04 20:58
Although a nice little argument, it's flawed for any chess playing entity but God, since it assumes that we know if a position is won, lost or drawn. Since we (usually) don't know this, we use heuristics to guide our game towards a hopefully winning position. To be able to compare and discuss these heuristics we use terms like "aggressiveness", and "positional". These are of course not very well defined terms, and people like vas and larry have to compare heuristics on a far more fine grained level. Nevertheless, the set of heuristics a chess playing entity uses to reach it's goal contitutes it's style, and any term used to describe them are imho valid.
Parent - - By 8lrr8 (***) Date 2008-01-05 03:21
"...and any term used to describe them are imho valid."

my argument was not about validity.  it was about such terms being of little value as they do not give us any meaningful measurement of playing strength.
Parent - - By Uly (Gold) Date 2008-01-05 03:34
But on two players rated 3200, I'm better off watching the games of the aggressive one than those from the solid one.
Parent - - By 8lrr8 (***) Date 2008-01-05 03:43
agreed, this is likely because the "aggressive" engine is likely playing sharper lines.  plus the fact that chess is mostly tactics (at least 80%?) anyway.
Parent - By Hetman (*****) Date 2008-01-05 19:51
Hi,

It depends what do you understand by tactics - if combinations I would have some doubts.
By using proper strategy the sharp play is to avoid,  it may cause that thouse 80%will have very little chance to happen.

Rgds
Hetman
Parent - - By Carl Bicknell (*****) Date 2008-01-05 18:17
VYTRON (& anyone else especially Vas!)
I've been thinking about what you've been saying above: bear with me on this.

1) What you're basically saying, I think, is that Rybka takes time to select a few good candidate moves. It then spends a very high proportion of its time checking that these moves are indeed good ones. If the top one is a good move it plays it, if not it goes to the second one and so on. Essentially, therefore, Rybka is checking that it has played a good (not necessarily BEST) move that doesn't do any harm.
IMPLICATION: Rybka is highly unlikely to play a bad move, but also poor at finding winning shots because once content with its top candidate move it doesn't devote much attention to alternatives.

2) By contrast you think that other chess progs search for the BEST move and if they find a win they play it, if not they search more proportionally through the list looking for the one special move.
IMPLICATION: They have a much higher chance of finding winning shots or "best moves", where they exist, but also a much higher chance of doing harm and playing a bad move.

So, therefore, Rybka works on the philosophy that if it consistantly plays GOOD positional moves it cannot lose, but doesn't go out of its way to search for the BEST move. This leads, therefore, to a very solid playing style.

Have I understood you correctly?
Parent - By Uly (Gold) Date 2008-01-05 23:00

> Have I understood you correctly?


Yes, but don't forget that this is only my opinion, and that maybe I'm totally wrong about it.

> Rybka is checking that it has played a good (not necessarily BEST) move that doesn't do any harm.


If by best, you mean a move that gives it the most chances of winning the game, then we can't know what this best move is unless we solve the game, so we never know if Rybka played it.

> They have a much higher chance of finding winning shots or "best moves", where they exist, but also a much higher chance of doing harm and playing a bad move.


I think that this is what Rybka Winfinder does, and it performs really bad compared to normal Rybka, so I can't really blame the programmers for staying with the solid approach.
Parent - - By Vasik Rajlich (Silver) Date 2008-01-06 14:25
I don't really think that this is correct. You always want to play better moves, this is a sort of axiom.

Rybka does tend to change her mind less often than other engines. It's a feature of her search, and this does hurt performance on tactical testsuites (although it's not the main issue there).

Vas
Parent - - By Carl Bicknell (*****) Date 2008-01-06 17:39
thanks for the reply....but you tease us!
won't you say anything about the fish's methods, or why we were wrong?
Parent - - By Vasik Rajlich (Silver) Date 2008-01-08 14:14
These things are complicated. Search and eval work is highly technical, and then at the end some sort of behavior just sort of emerges. It's just my general observation that there is no real difference between trying to play good moves and trying to play best moves.

Vas
Parent - - By Uly (Gold) Date 2008-01-08 21:27

> It's just my general observation that there is no real difference between trying to play good moves and trying to play best moves.


But I think there is. Best moves are harder to find, so you should either spend time finding those (and if you don't find them in time, you may end playing a bad move) or finding a good move. Good moves are easier to find, so easy that Rybka finds a good move in the first iterations, and then just spends her time getting sure that they're really good. The success of Rybka would then be that she's the best at finding good moves (she's solid).

You could argue that these good moves have more chance to become best, so by finding good moves you're already searching for best ones (And probably you're right), but I think that most of the time, best moves look bad at the beginning, and only after really deep search they prove best, but since you don't know how to know what these best moves are, you'd need to spend lots of time figuring that out, time that maybe you don't have, and you end without a best move and without a good move. The failure of other engines would then be that they spend too much time trying to find a best move and are slow at finding good ones (they're speculative).

Good moves aren't necessarily aggressive, and it could be the reason that Rybka isn't aggressive at the beginning (And if the opponent never plays a mistake, Rybka will stay like this.)

My solution would be to let Rybka use half of her time to find a good move, and then (now that you have a good move) become some sort of Winfinder to check if you're missing some deep win on the other moves. But this wouldn't work very well against older Rybka versions, so maybe having an "Anti-Old-Rybka" (the default personality) and an "Anti-Computer" (One with a different behavior that knows that the opponent is a weaker computer and tries to exploit it) separated personalities would be best (Since we already have the Anti-human one, by contempt.)
Parent - - By turbojuice1122 (Gold) Date 2008-01-08 22:21
I think that the "best moves" of which you speak are primarily extremely difficult to find in highly tactical positions.  In tournament time-control games with Rybka vs. Rybka, if God was to provide analysis, I don't think that we'd see a great amount of shockers.  However, in games with Junior vs. Junior at the same level (i.e. much, much longer time controls than Rybka vs. Rybka, such that we're nearing correspondence time controls), God would probably show us a whole lot of real doozies that would leave us scratching our heads for years.
Parent - - By Uly (Gold) Date 2008-01-08 23:35

> if God was to provide analysis, I don't think that we'd see a great amount of shockers.


Once Rybka 4 appears and shows you how bad 2.3.2a really plays, your opinion will change, and you will use the same argument with Rybka 4 Vs. Rybka 4 as an example.
Parent - By 8lrr8 (***) Date 2008-01-09 08:09
i think what we'll see is a marked increase in strength at short time controls.  so rybka 4 may be radically stronger than 2.3.2a at say, 40/40.  but making engines stronger at 120/40 is gonna be really tough from here on out.  the fact that we have ultra-strong book authors doesnt make this any easier.
Parent - - By Kapaun (****) Date 2008-01-08 22:41
Well, I don't know. Playing good moves is just not good enough nowadays. At least not in correspondence chess, and Rybka is quite a champ there, too. Also, I don't think that best moves would necessarily look bad at the beginning. I think that most best moves just look fine anyway, and that Rybka will find them more often than not. However, even that's not enough for correspondence chess, and so you need to do some additional analysis, trying to find out those cases, where Rybka's choice ist not the best thing to do...
Parent - - By Uly (Gold) Date 2008-01-08 23:37
Rybka is not playing best moves most of the time, that's why Centaur + Rybka > Rybka, because the human is better at finding best moves with Rybka's help than Rybka alone.
Parent - - By turbojuice1122 (Gold) Date 2008-01-09 03:34
But still, it's rare for any of these moves to be real shockers--they're ones that you look at and say, oh yeah, that is a pretty good move.
Parent - By Kapaun (****) Date 2008-01-09 10:06
That's right. Already with the early Rybka versions my impression was that her big strength was not winning more game than any other engine, but losing less games than any other engine, and with a margin...
Parent - By 8lrr8 (***) Date 2008-01-09 08:13
of course, altho freestyle tourney use 90min time controls.

the gap in strength btwn centaur+rybka vs. stand-alone rybka narrows considerably at classical time controls.  especially if it's an vapor-chilled, OC'd 8-core rybka w/ a super opening book.
Parent - By Kapaun (****) Date 2008-01-09 10:02 Edited 2008-01-09 10:10
That doesn't invalidate what I said. In order to change the course of a game you need just one move. Give me a move, Vasily. One move only, please. :-)

Rybka may find the best possible move 49 times out of 50 (which is not the case, not even close), and still a centaur may win.

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill