Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Jealousy attacks against Vas, Larry & Rybka 3 on CCC forum
- - By Silvian (***) Date 2008-08-11 13:12 Edited 2008-08-11 13:15
Parent - - By Arkansaw (***) Date 2008-08-11 13:20 Edited 2008-08-11 13:24
I think his posts are acceptable, some people are just too jumpy, and that's how internet flame wars begin
Parent - - By Sesse (****) Date 2008-08-11 15:45
I really cannot see what the problem here is either. People disagree with Rybka's node counting method; there is some speculation about why, but hardly anything I'd call "jealousy attacks".

/* Steinar */
Parent - - By Silvian (***) Date 2008-08-11 17:32 Edited 2008-08-11 17:44
Really ?

Take a look to this post:

http://64.68.157.89/forum/viewtopic.php?topic_view=threads&p=207235&t=22900

"Rybka is IMHO a brutal effective and fast searcher. It is as dumb as Toga for example. It has very little knowledge about chess."

I think-now-you can :)see :) the problem !

Regards,
Silvian
Parent - - By Sesse (****) Date 2008-08-11 19:40
"Jealousy attacks" on Vas, Larry and Rybka? Because he thinks (rightfully or not) Rybka is a fast searcher with little knowledge? Grow up. :-)

/* Steinar */
Parent - - By Arkansaw (***) Date 2008-08-11 22:18
Actually, who knows unless you have the source code?
Parent - - By Uri Blass (*****) Date 2008-08-11 22:23
No need to have the source code to know that a program that wins the other programs has a lot of knowledge.
Toga also has more knowledge than most programs otherwise it could not beat most programs.

Size of evaluation is unimportant because it is the quality that is important and not the size.

Uri
Parent - By Arkansaw (***) Date 2008-08-11 22:37
That is disputable because results != knowledge.....sometimes it's just a matter of picking the right battles to get the results you want
Parent - By theoak (**) Date 2008-08-11 23:01
It is hard for me to believe that the best team in terms on chess experts and programmers can produce a machine with little chess knowledge!!! I don't think that only Vas is responsible for Rybka. I also don't see these remarks as attack. They are rather random and not based on evidence.
Parent - By Kapaun (****) Date 2008-08-11 23:13
I don't. What exactly would this problem be?
Parent - By grolich (***) Date 2008-08-12 04:58 Edited 2008-08-12 05:15
That more or less says that Rybka's evaluation doesn't have a lot of knowledge.

The only person who has any REAL right to be offended by this is Larry.
After all, it basically says his work on Rybka has not been much help, and that almost everything concerning Rybka's success is search related

As long as Larry decides to be the grown man that he is and not get into flame wars (or just not be offended by what every person in the world thinks of his work), I see no reason why should we, who have a somewhat lesser reason to be offended (*understatement of the year*),  be doing that.

Just let it be. There will always be people who do that.

Besides, a quick read of that last post you linked to should pretty much convince anyone that the poster is not someone to be taken seriously.
Giving a single position that is evaluated wrong as proof (or very strong support) of his claim that the program has little knowledge (his own words: "that says it all").... Not a very scientific thing to do.
(you have noticed that that last poster is NOT Hyatt, I hope...
Hyatt's posts are more moderate. He has his opinions, whether he really thinks he has reasons for his "suspicions" or it is just jealousy (or both) is not really our business. He's entitled to his opinion).

why take everyone who has a negative opinion so seriously?
Parent - By Quapsel (****) Date 2008-08-12 07:50
In what a sheltered environment did you have grown?

Here I can see no problem.
Probably he is not right.
OK, then he made a mistake.
So what. (maybe he speaks so, allthough he knows it better. Maybe he really thinks so.
But this 'Rybkas search is faster, it is not her chess knowledge!'-thesis is pronounced by some people.
Probably this is wrong. So what.)

The world is not pink dollhouse, where all men are free of mistakes and speak honestly each time.

Quap
But smiling about the 'wrong rook pawn'-indignation.
the 'Rybka is so very, very expensive'-outrange
and the 'Rybka makes so very, very stupid mistakes'-postings.
No, not everybody likes the strongness of Rybka and the success of the Rybka-team.
Go on!
Parent - By Nelson Hernandez (Gold) Date 2008-08-11 14:04
Water off a duck's back.  If they were all rending their garments in rage, and were accusing Vas of having fathered illegitimate babies or committed infamous grand larcenies, it wouldn't matter.

Rybka 3 speaks for itself.
Parent - - By Carl Bicknell (*****) Date 2008-08-11 14:44
It is pretty clear what Hyatt is saying.
He is saying that Rybka is a very fast low-knowledge program pretending to be a slow high-knowledge program.
Parent - - By BB (****) Date 2008-08-11 15:04

>It is pretty clear what Hyatt is saying.
>He is saying that Rybka is a very fast low-knowledge program pretending to be a slow high-knowledge program.


Did I miss something? I read all of the posts of Hyatt (except some of the exchange with Rolf), and he never claimed that about R3 (only in an allusion to Strelka here).
Parent - - By Carl Bicknell (*****) Date 2008-08-11 15:13
It is clearly implied. He didn't use those words but that's the effect of all his posts taken together. I think one of the people he exchanges with on the forum said it directly. (I presume he is referring to Rybka 1/2.3.2.a because to call R3 low-knowledge is silly)
Parent - - By BB (****) Date 2008-08-12 04:35 Edited 2008-08-12 04:40

>>It is pretty clear what Hyatt is saying.
>He is saying that Rybka is a very fast low-knowledge program pretending to be a slow high-knowledge program.
> [...] He didn't use those words but that's the effect of all his posts taken together. I[...] (I presume he is referring to Rybka 1/2.3.2.a because to call R3 low-knowledge is silly)


If he is saying that about R1 (or even R2.3.2a), I cannot say that I disagree. However, "pretense" can be a psychological observer effect - for instance, Ed Schröder was convinced that Rybka had made great gains via eval rather than search, whilst others (most notably Christophe Théron and a certain Dagh Nielsen) preferred to think that Rybka was a fast/efficient searcher.
Parent - - By Uri Blass (*****) Date 2008-08-12 05:31
unlike you I disagree about the low knowledge part.

1)knowing which lines to search is knowledge
2)quality is important and not quantity.
A small evaluation does not mean not having knowledge and I believe that the evaluation of toga ot strelka is better than
the evaluation of many chess programs with bigger evaluation.

I think that rybka is a fast high knowledge program.

Uri
Parent - - By Carl Bicknell (*****) Date 2008-08-12 07:19
"I think that rybka is a fast high knowledge program."

how would you classify the other well known engines?
Parent - By Uri Blass (*****) Date 2008-08-12 07:39
I do not know but I think that the main advantage of rybka is not speed.

Speed can explain advantage of 2:1 but not advantage of at least 6:1
The main advantage of rybka is knowledge.
It is probabely both better evaluation knowledge and better knowledge which lines to search.

Uri
Parent - - By diskamyl (**) Date 2008-08-11 20:00
Well it's kind of obvious:

"And if you happen upon an idea that is unique, but not that hard to understand, then you could easily be concerned that with so many eyes looking on, the more information you provide, the easier it becomes to figure out the new idea. It is much easier to hide behind a curtain, and say "my NPS is much lower because I have so much knowledge in my program" and let that throw people off the track. Until someone actually disassembles your program and posts the details, including the fact that it is _not_ a knowledge-based evaluation-heavy program at all. Smile"

http://64.68.157.89/forum/viewtopic.php?t=22900&postdays=0&postorder=asc&topic_view=flat&start=10

Some people have no shame. Vas strictly made it clear that NPS were lower because they were simply calculated differently. I never remember him saying something even close to "it's low because it has so much knowledege".

and again: Rybka 3 - Crafty:  50 - 0. :)
Parent - By Arkansaw (***) Date 2008-08-11 22:20
I don't think logical arguments can be replaced with personal attacks. It's basically one man's word against another unless you know exactly how the calculation is being done
Parent - By BB (****) Date 2008-08-12 04:28

>Vas strictly made it clear that NPS were lower because they were simply calculated differently.


This is precisely Hyatt's point (that he made like 5 or 10 times in those posts) - the idea of "differently calculated" NPS is, for the most part, a fiction.

Nodes are nodes are nodes are nodes, except possibly for minor details about legal moves, hash hits, et cetera. Rybka 1.0 Beta counted something (claimed to be essentially 8 times the number of calls to trans_max_store from full_search) other than nodes and reported that number with the "nodes" moniker (at one point, I suggested calling them shtuka rather than nodes - and I also blamed the UCI interface for not allowing variant words for "nodes" :)). There is no obvious mapping between these "entity counts" and nodes, though one can give a generic approximation.

Later Rybka versions essentially count calls to make_move_white, and then divide by 7 - if this division were instead a multiplication by two, you'd get a reasonable "node count" in the informatics sense of the word. As indicated in a different post, the obvious solution is simply for the GUI to multiply the Rybka "node counts" by a suitable multiplier, which would be 14.
Parent - By Roland Rösler (****) Date 2008-08-12 07:55
So what? She is the best! Who cares about slow and fast or low-knowledge and high-knowledge. They have to do better.

PS: I´m slow with high-knowledge! :-) If I could become better with fast and low-knowledge, I wouldn´t choose it. But I´m no engine.
Parent - By DeletedAccount (**) Date 2008-08-11 19:34
Full of hate. But why bother? It is his own time he wastes.
Parent - - By hiarcs1 (*) Date 2008-08-12 01:31
In my opinion I think that a chess engine will all the best search techniques and tricks but with minimal knowledge will resemble to play like it has a lot of knowledge.
I conjecture the following:

The limit of a chess engine's chess knowledge as the search goes to infinity is the solution to chess.
This means that, with only basic evaluation, an engine will play like it has a lot of chess knowledge if it has sufficient search speed and width.
Parent - By nine castles (**) Date 2008-08-12 01:44
This is correct, but it's important to consider that the evaluation has a serious impact on the search. Given that no engine will ever have perfect information (or else it would not really be searching), the engine must have a meaningful evaluation function in order to direct its search toward more interesting lines.

For that reason I don't think it makes sense to clearly distinguish between "Search" and "Eval" since they are blended together at many steps.
Parent - - By tomski1981 (*****) Date 2008-08-12 02:30
in my opinion: good search + poor knowledge = engine that draws a lot (naum?)
Parent - - By Quapsel (****) Date 2008-08-12 08:10
For wich reason do you think so?
"good search + poor knowledge" means:
The program will find / pretend more positions, that are in its horizon.
It sometimes will not decide, that the situation is bad, all positions behind the horizon semm to be bad.

This will give several wins because of deeper search!
And  because of small knowledge it can go into  Situations where 'all is bad'.
And it will loose.

Ich don't think, that such Engines are more drawish than other.
A 'good knowledge'-Engine maybe could try to hold positions that are drawish.
A 'bad knowledge'-Engine has to walk though a minefield and has to be aware, that the opponen each time can try to jostle him to a bomb. Maybe he sees the mines, but the mines are all around!

Quap
Parent - By tomski1981 (*****) Date 2008-08-12 09:45
it takes more knowledge to find a win than to hold a draw... an engine with poor knowledge might be able to find draws faster, due to outsearching, but often won't be able to convert to a win, because although it can search deep, it doesn't follow correct lines.
Parent - By InspectorGadget (*****) Date 2008-08-12 05:38
Little knowledge, poor endgame, dumb but beats everybody up :)
Parent - By Zarkon (***) Date 2008-08-12 06:17 Edited 2008-08-12 06:20
"Jealousy attacks"? What a load of nonsense . I agree with "grow up"!
Up Topic Rybka Support & Discussion / Rybka Discussion / Jealousy attacks against Vas, Larry & Rybka 3 on CCC forum

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill