Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / More threat to Rybka now ?!
- - By Jouni (**) Date 2007-12-10 14:21
Parent - - By turbojuice1122 (Gold) Date 2007-12-10 14:51
The names sound like they were pulled from the movie, "Idiocracy".
Parent - By RFK (Gold) Date 2007-12-10 16:04 Edited 2007-12-10 16:10
That is extremely sad! I am not seeing any of those engines showing up on Ridderkirks listing or Superchessengine.com.
Parent - - By InspectorGadget (*****) Date 2007-12-11 07:46
I won't install an engine with such a name in my PC. Who would want to give his engine such a name?
Parent - By Uly (Gold) Date 2007-12-12 08:51
What if this engine was stronger than Rybka?

If an engine is stronger than Rybka, I don't care about the name, it could be called "Vytronfailsbadly" and I wouldn't care about its name.

(That's not the case with these engines, but I was talking hypothetically)
Parent - By noctiferus (***) Date 2007-12-10 15:58
winboards?
Parent - - By noctiferus (***) Date 2007-12-10 21:28
I've run a very short gauntlet, 4' 2'' / 4 games match, HS_book, with toga131 as benchmark, just to have a feeling on strenghts (for what is worth)...
not enthusiasmating
Parent - - By Felix Kling (Gold) Date 2007-12-11 22:31
You have to build a rocket engine which combines fucka and mutha, this is the ultimate engine. Testing those engines seperatly doesn't make sense.
Parent - By billyraybar (***) Date 2007-12-12 15:54
lol
Parent - By noctiferus (***) Date 2007-12-12 17:02
a bit too hard for me! :)
Parent - - By Shaun (****) Date 2007-12-12 17:27 Edited 2007-12-12 17:52
See here

http://schachwerkstatt.foren-city.de/topic,302,-grossangriff-auf-rybka.html

I believe Christopher in misguided, I am not sure how you stop people cloning or pirating engines but this seems counter productive to me and just wastes everyones time?

I will make just one comment regarding CCRL and Strelka as it is mentioned in the tread...

I think Strelka is probably* a clone (depending on you definition) and CCRL are unlikely to run any more games with the engine unless it is for comparison purposes, our site provides correlation statistics showing how similar Strelka is to other engines - it would seem counter productive to remove this information from the public arena. However Strelka has effectively been removed from all rating lists and can only be found on the complete list.

* there are still a number of computer chess experts that claim Strelka is not a clone and that is why I say probably.

P.S. If you see a new clone or potential clone tested at CCRL please let us know - our forum can be found here: http://kirr.homeunix.org/chess/discussion-board/viewforum.php?f=7
Parent - - By Vasik Rajlich (Silver) Date 2007-12-13 12:25
Shaun,

your correlation statistics are very interesting and I'm a big fan of them, but they are useless for clone classifications. It's easy to take the source code, change some constants, and get completely different behavior.

There are three ways to classify something as a clone:

1) Executable comparison
2) Source code comparison
3) Author confession

It's very rare to have all three, as we have here.

> * there are still a number of computer chess experts that claim Strelka is not a clone


This is not correct.

Vas
Parent - - By Debaser (***) Date 2007-12-13 13:01
On CCRL 40/40 ponder hit and eval diff for fritz 11 and 10 vs Rybka 2.3.2a 64-bit are:

Fritz 11 68.3% 0.28

Fritz 10 60.5% 0.59

What does this mean? Is this suspicious? or just that as Rybka is "The Goddess of Chess" if you want to improve, you have to be like her, but there are several ways to arrive to the same place.
Parent - - By turbojuice1122 (Gold) Date 2007-12-13 13:32
Okay, so I didn't really think this was significant when Pia mentioned a couple of examples yesterday between Fritz 11 and Rybka 2.3.2a.  I am now starting to think that this is worthy of thorough investigation.  One of the big things about Strelka and Rybka 1.0 is not only that they have such strong ponder-hit and evaluation correlations, but because of the trends that exist: Strelka shares this with no other program except for Rybka, while Loop, which also has a strong correlation, shares this with lots of programs, particularly ones in the Fruit family.  I seem unable to find the general statistics for Fritz 11, but I would guess that Fritz 11 shares such a high correlation with no other program except for Rybka.  The next step would be for people with more time than I have to try the two programs in an array of test positions, particularly some "strange" ones like what Uri had before, and particularly ones where Rybka 2.3.2a shows a bug.  Also keep in mind that at tournament time controls, Fritz 11 is about as strong as Zappa Mexico and Shredder 11, so one cannot just say that you have to be like Rybka to improve.  However, what Pia showed yesterday was actually an indication that there is not suspicious behavior: the evaluation profiles were different (while with Strelka, they're the same).  The situation would be far more difficult if the very astounding idea that Fritz 11 might contain parts of Rybka is true: Fritz 11 has been developed on its own and is a far different program in many respects (which is why this correlation is so very odd), and so finding exactly what Fritz 11 took will be far more difficult than it would be for Strelka.

Anyway, at this point, I'm putting money on that everything about Fritz 11 is legitimate, but some of this is due to my faith that the Fritz team is morally refined and wouldn't do something untoward in this realm.
Parent - - By Debaser (***) Date 2007-12-16 17:15 Edited 2007-12-16 17:18
Hi turbojuice1122.

I do not think that you can handle this in the same way that Rybka-Strelka. Fritz engine has existed a lot of years before Rybka appeared. And Strelka appeared after Rybka. So trying strange positions or bugs should not work, I have not tried though ;)

So we can not blame Fritz team of cloning, I understand cloning here as using complete or significant parts of other engine. Other thing is that you can use "a part" of other engine, in this case evaluation is quite similar now, is suspicious then or they have just learned from Rybka play?

Can any expert tell me if the changes in Fritz on CCRL have a normal explanation?
Parent - By turbojuice1122 (Gold) Date 2007-12-16 18:51
I really think that the only reasonable explanation in the case of Fritz 11 is that the Fritz team has finally on their own stumbled upon some of the techniques that Vas has been using for quite awhile; I think they would have too much at stake to get involved in something like cloning.  I don't think that they are able to use "parts" of Rybka without actually getting into Rybka itself--the logical parameters involved in a chess program just seem too vast and complicated for anyone to "learn from Rybka play" and be able to have matching evals.

The main difference that I can see between Fritz 11 and Strelka in their behavior is that in terms of the evaluation profile, Strelka often perfectly mimics Rybka's play; the reason for this is obvious, of course, as Vas notes below: he has Strelkas source code, and knows that Strelka is a Rybka clone.  In the case of Fritz 11, on the other hand, even though the actual numbers match, the profile itself does not: Fritz 11 will reach a local maximum at a different depth than Rybka and around local maxima and minima will have different fluctuations than Rybka--this is what I mean about Fritz 11 having a different profile, while Strelka has the same profile.
Parent - - By Vasik Rajlich (Silver) Date 2007-12-16 16:05
It's just part of the game. It's up to me to protect Rybka's algorithms ..

Vas
Parent - - By Debaser (***) Date 2007-12-16 17:08
You have left me with the some doubts I had before asking ;)
Parent - - By Banned for Life (Gold) Date 2007-12-16 18:06
If another chess program used the exact same algorithms as Rybka uses, and the efficiency of the two programs was always equal, it would be impossible to distinguish them by their playing style even if they didn't share any code.

Since Rybka is top dog, it is natural that other engine developers want to emulate her algorithms. To the extent they are successful, their engines will be stronger and play in a manner more similar to Rybka. This is not, in and of itself, a bad thing. Most of the algorithms used by Rybka were around before Rybka appeared and Vas made his contribution, whatever that is.

From a legality standpoint in the US, copyright could be violated if portions of the code are copied (very straightforward with lots of established precedents), or if the program is substantially the same as an existing work (much more nebulous with few established precedents). Apparently Strelka would be in violation in both cases.

Reverse engineering to pull out algorithms has also been a major battleground in the US, and most of the precedents have supported the people doing the reverse engineering. Pulling out the underlying algorithms in this manner would be a non-trivial task, but its always a possibility.

Of course from a practical standpoint, none of this is really relevant because there is far too little money in chess to mount any sort of legal challenge under almost any conceivable circumstances (well, maybe if someone renamed a recent Rybka release and started selling it).

Alan
Parent - By Vasik Rajlich (Silver) Date 2007-12-20 13:22
Yes, this is a pretty good summary.

Practically speaking, as a developer, you want to make your algorithms more difficult to pull out. This is actually a fun game - it's a little bit like 'Home Alone' :)

Vas
Parent - - By turbojuice1122 (Gold) Date 2007-12-13 13:37
There are three ways to classify something as a clone:

1) Executable comparison
2) Source code comparison
3) Author confession

It's very rare to have all three, as we have here.


This seems to be the bombshell that many of the skeptics have been looking for, so I think that it would be best now for everyone if you could elaborate a bit on all three of these and why we have these confirmed with Strelka.  On (1), without giving away too many of your methods, how can you tell that the executables are the same?  On (2), it has been stated before by others that you had not looked at Strelka's source code: have these statements been false, or have you now rectified this?  On (3), to which particular statement of Yuri's are you referring, and how does this constitute a confession?  I know that you have talked somewhat on (3) in other posts, but it would seem good to have it here, especially since there has been debate whether what he has said actually constituted a confession or not.  As far as we can tell, all that he has said is that he basically made the weights of Strelka equal to the weights of Rybka by detailed analysis of Rybka's evaluations in many positions and such.
Parent - - By Vasik Rajlich (Silver) Date 2007-12-16 16:09
Yes, I have the Strelka sources. It's just an obvious situation.

Vas
Parent - - By GenoM (**) Date 2007-12-22 10:28
It seems you finally received the sources of Strelka.
We're waiting for some kind of more concrete statement about this case.
Parent - - By Dadi Jonsson (Silver) Date 2007-12-22 15:50
Who are the "We" that are waiting?
Parent - By GenoM (**) Date 2007-12-22 22:34
Me and group of russian scientists.
Up Topic Rybka Support & Discussion / Rybka Discussion / More threat to Rybka now ?!

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill