Not logged inRybka Chess Community Forum
Previous Next Up Topic Rybka Support & Discussion / Rybka Support / Rybka 4 FAQ (18773 hits)
1 2 3 4 5 Previous Next  
Parent - By Akbarfan (***) [de] Date 2010-06-04 12:04
Yes, I think the same.

Playing blitz during the Beta phase it's not the same as serious analysis.
- - By tonee [us] Date 2010-06-17 17:14
i have the chessbase version of rybka 4 and the opening book, but the opening book wont copy to the hard drive. also the opening book has averages that are 200 elo pts below rybka 3,whats going on ?
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-06-18 08:05

> also the opening book has averages that are 200 elo pts below rybka 3,whats going on ?

I guess Elos can only be compared within each book.

Parent - By yanquis1972 (****) [us] Date 2010-06-20 00:47
i believe the elo's are based on some combination of human otb/correspondence/playchess ratings.  all these pools will average much lower than a hypothetical rybka 4 fide rating.
- - By fisherino (*) [us] Date 2010-06-19 15:19
For some reason i am having trouble installing Deep Rybka 4. Any suggestions would be appreciated.
Parent - - By Uly (Gold) [mx] Date 2010-06-19 16:28
What problem are you having?
Parent - By fisherino (*) [us] Date 2010-06-19 23:42
I dont know how copy the chess engine to a directory on my computer which is step one in the installation in Chessbase 9 (Chessbase 10)
- - By Beach [de] Date 2010-06-20 18:14

why stop DeepRybka 64-UCI-Engine (Download-Version from convekta) analysis on deep 18-21 (under Chessbase10/Fritz12/Shredder-GUI) ?

Best regards
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-06-21 07:38
I will write more about this issue in the next week or two.

Parent - By Beach [de] Date 2010-07-05 08:56
Have you a solution now for this problem? It is impossible to use DeepRybka4 for a deep analysis.

Best regards
Parent - - By Beach [de] Date 2010-07-13 18:09
Why don't you answer me to this big bug in DeepRybka 4 for about three weeks ???

Best regrads
Parent - - By keoki010 (*****) [us] Date 2010-07-13 20:23
I think he and his wife are busy having a baby. :wink:
Parent - By Beach [de] Date 2010-07-14 22:29
Ok, thank you for your hint.

Best regards
- - By sem (*) [me] Date 2010-06-21 20:03
If you can explain to me what is the difference between Deep Rybka 4960 x64, Deep Rybka 4 SSE42 x64, and Deep Rybka 4 x 64
Which is the strongest engine in blitz game?   I use 64-bit Windows 7
Parent - - By Uly (Gold) [mx] Date 2010-06-21 21:16

> Which is the strongest engine in blitz game?

SSE42 if your processor can run it. 960 version plays Chess960.
Parent - - By sem (*) [me] Date 2010-06-22 09:03
My processor is AMD ATHLON(tm) II X3 425 2.70 GH. Is it ok?
Parent - By Fulcrum2000 (****) [nl] Date 2010-06-22 11:15
According to the dump found here it supports SSE4a, so yes it should work for you.
- - By eddy158k Date 2010-06-23 03:39
Hi everybody,

Can someone tell me Where to find the Book Options on R4 ?
Parent - - By Uly (Gold) [mx] Date 2010-06-23 04:29
What interface?
Parent - - By eddy158k Date 2010-06-23 13:55
I suppose is Fritz. I just install this Rybka 4 but I haven't install Fritz yet. If I need to do it let me know where in Fritz can I locate the Book Options. Thanks in anticipation.
Parent - By eddy158k Date 2010-06-24 15:38
Can anyone else give me an advise here as to where can I find the Opening Book Options on Rybka 4 without interface or if I must install one to have access to it ?
- - By pwbt72 (*) [us] Date 2010-07-21 00:52
How to stop those annoying tech msgs upon starting R4.????
Seeme to have missed it somehwere.
Everyone knows this but me.
Parent - - By Uly (Gold) [mx] Date 2010-07-21 01:04
What annoying tech msgs?
Parent - - By pwbt72 (*) [us] Date 2010-07-23 05:01
The one that appears upon loading R4  Children restarted
                                                     Hash size set to 128mb
                                                     nalimov path set to.......
                                                     etc usually a repeat
Sorry I was not clearer.
I did read about this somewhere once but failed to remember
Thanks on advance
Parent - - By Uly (Gold) [mx] Date 2010-07-23 09:04
I don't know if there's a way to disable them, sorry.
Parent - By Lukas Cimiotti (*****) [de] Date 2010-07-23 20:05
Yes, you have to use a different GUI. It's only the old ones that display this.
- - By natmaku (**) [sa] Date 2010-07-22 11:34
What is changed in the parameters for "Deep Rybka 4 Beta15"?..Thanks!
Parent - - By Uly (Gold) [mx] Date 2010-07-22 12:01
Beta15 settings: Normal Move Time 100, Max Move Time 100, White Pawn: 3, Black Pawn: 3, White Knight: 9, Black Knight: 9, White King Bishop: 9, Black King Bishop: 9, White Queen Bishop: 9, Black Queen Bishop: 9, White Rook: 15, Black Rook: 15, White Queen: 30, Black Queen: 30.
Parent - - By natmaku (**) [sa] Date 2010-07-22 12:20
thanks a lot!!!
Parent - - By Uly (Gold) [mx] Date 2010-07-22 12:39
Please note that the reason those settings over-perform remain unexplained, they shouldn't, it's just a random buggy version that all testers had performing worse than the release version.

The ongoing theory is they were just lucky (very, very lucky since they remained on top even after much more games were added, and performed just as good as the best Rybka 4 time control settings last time I checked).
Parent - - By Alkelele (***) Date 2010-07-22 20:20
I think it is down to the law of randomness. Let's say you create 10 versions of Rybka 4 with slightly different time control parameters or piece value parameters. Changes so minor that you know that they are insignificant.

Then you go ahead and create a "test" of these versions, and you will see one version get out on top.

Does that then mean that this version is the strongest version?

No, I think it just means that you will never see a field of equal competitors match each other exactly with equal scores. There is some randomness involved.

I think that when you have these adjustable parameters, and you go ahead and make "tests" (I put in quotation marks since they are not extremely exhaustive tests), then you will always initially have some version or another outperforming the field.

But, in the next tournament/test, it might as well be another version that is doing better.

So, my suspicion is, whatever "optimal" or "default" Rybka 4 parameters is settled on, one could always tweak them a little in 10 different ways, and then create a tournament with these 10 tweaked versions plus Rybka Default, and then odds are that Rybka Default would not win in the next test. But this doesn't mean that Rybka Default is inferior, it just means that it is outnumbered 1 to 10.

Something similar was happening back in the Hydra period, when people were suggesting that Rybka was (already) stronger. But, Hydra was playing only as one entity, but it was up against 50 Rybkas or whater in the Freestyle tournaments. Of course, odds then were that one Rybka or another would outperform it in a Swiss format. Try to imagine the opposite situation, with 50 Hydras vs. 1 Rybka. Rybka getting into the final would surely be a great achievement then. The point is, it is always very difficult for the favorite or the best version to truly perform the best in every test.
Parent - By Uly (Gold) [mx] Date 2010-07-23 03:16
Indeed! I recall someone actually went and did a similar test to what you're proposing, getting strong evidence that Rybka 3 is 50 elo stronger than Rybka 3 :smile:
Parent - - By Nelson Hernandez (Bronze) [us] Date 2010-08-03 11:58
I agree with everything you say except its general yet unspoken conclusion, which is that nothing is knowable through testing.  Of course you can form conclusions.  You merely need to manipulate one variable and run a series of lengthy tests, and you need rigorous statistical analysis.  After a while you should see a performance curve emerge where moving the setting too much in either direction results in obviously inferior results, and from that you can infer what is the happy medium.  It's painstaking work.

Next you manipulate a different variable and repeat.  Yes, of course changing this second variable very probably influences the first, but the general idea is to tweak one variable after another and then compare against your original set of variables.  This approach ought to make slow, incremental progress over time.

This is similar to the technique that has traditionally been used to develop new drugs.  Of course, it takes considerable computer resources, time and discipline to develop something better, but it can be done if you know how to do it.
Parent - - By Banned for Life (Silver) Date 2010-08-03 13:45
There are two problems with the optimization method you propose which cause it to fail in most practical cases:

1) The features are highly correlated, i.e. they have very significant mutual coupling factors. While this is true, you can't increase the weight of one feature without increasing and decreasing the weights of many other features, and
2) Even if the features are independent (or at least decorrelated in the nonlinear case), most complex functions will have local maxima where your method will get stuck.

The solution to both of these issues is to modify all of the weights simultaneously and look for improvements. This solves the first issue directly, and solves the second issue by allowing contributions from certain features to fall if there is a greater increase from other features.

The main problem with any technique though is the time required to perform a statistically significant test. I've suggested to Vas that instead of performing A/B tests using a new set of feature weights vs the default and running a huge number of games, that he consider picking out a few million positions and evaluating them for a second each, and then using these as truth values and modifying the result of static evaluation using the perturbed feature weights to approximate the 1-second ground truth as closely as possible.

If you accept the assumptions that:

1) A 1-second eval will be considerably better than the static eval in most cases, and
2) A few million positions will be enough to cover most evaluation requirements,

this method provides a method of quickly improving feature weights without running any games.
Parent - - By Nelson Hernandez (Bronze) [us] Date 2010-08-03 14:13
An amateur meets a professional!  You are devilishly smart, Alan.  I don't know why Vas doesn't hire you as a consultant.  You could probably add 100 ELO all by yourself by completely revamping this kind of testing along more rational lines.
Parent - - By Uly (Gold) [mx] Date 2010-08-03 18:25
One thing I'm sure about is that Rybka 4 would have been a much better product have you guys been on the "beta" testing team.
Parent - By Nelson Hernandez (Bronze) [us] Date 2010-08-03 21:58
We'll never be insiders.  Besides, we're much happier on the outside with our sniper rifles.
Parent - By Banned for Life (Silver) Date 2010-08-03 13:28
The problems associated with testing a large number of hypotheses has of course been thoroughly investigated in other disciplines. In detection theory, it is called m-ary loss and one must use the number of hypotheses that will be looked at to calculate an increase in SNR required to maintain the desired statistical significance for the group of hypotheses when comparing them to a single hypothesis (e.g. the default).
- - By peterfabian Date 2010-08-03 03:59
hey i'm new here and i have Aquarium + Rybka 4 Download version. I have rybka 4 openning book as well installed under Windows 7 32 bits home premiun edition on a 2.0 Ghz dual core CPU. Everything it seems to be fine, i'm impressed for the level of the program, very strong, however i am very impressed also about why the program takes so long in playing when it founds a mate line (even mate in 1!) the other day it took almost 1 minute in making the move. Why is that? i have had other more modest and oldest program which played almost immediately when they found a mate why Rybka takes so long time for this? someone else has had this problem also? is there an explanation for this? (if this it is an issue, is there a patch for solving it?) sorry if this have asked before...thank you for your support...
Parent - - By peterfabian Date 2010-08-04 02:39
nobody has had this problem???:roll:
- - By Ron Bateman (**) [us] Date 2010-08-07 06:41
I can load the Rybka 4SSE42X64 engine into the Aquarium UCI, along with Rybka 4x64 an Rybka 4960, but it is inactive.  How can I activate it?  I also do not get it's parameter box.  Thanks.
Parent - By Banned for Life (Silver) Date 2010-08-07 06:45
You need a processor that supports SSE 4.2 to use this engine.
- - By CarlHeineman [us] Date 2010-12-21 22:40
Have a problem with the update downloader.  The product, Rybka 4 Aquarium, is registered by wheb I go to do upgrades, the program tells me I do not have enough rights.  What does this mean? and how do I fix it if it is a fixable problem?
Parent - - By nebulus (***) [no] Date 2010-12-21 23:47
I don't know how much Downloader is used these days, but they post updates in Aquarium sub-forum and on their official site.

4.0.5 update
4.0.6 update (public beta)
Parent - By Felix Kling (Gold) [de] Date 2010-12-22 03:43
actually many users are confused about this. the downloader isn't used that much by Aquarium anymore.
- - By MrSpock4 (*) [us] Date 2010-12-22 21:55
questions for rybka 4 chessbase (fritz)...

1.   the instructions to add jiri's opening book for rybka 4 do not work for me-- missing tabs (edit).  i would like some clear instruction for installing that book to rybka 4.  i've already done it for my rybka 3 program (aquarium) aquarium is much easier to use IMO and i'm used to it.  had no problem loading that book as well as some other ctg's.

2. Is i possible to run rybka 4 (chessbase purchase) to my 3.2.1. version of rybka w/aquarium?  i loaded it, but when i play the program, as soon as it runs out of book moves listed in the tree (no info) it stops playing.

3. is there anyway to rig up the online nalimovs (the button at the top left in analysis ribbon of aquarium) so the program can access them while playing a game, or is that just an analysis feature?

i have more questions, but these are for now.  thx everyone--
Parent - By DuWayneL [us] Date 2010-12-28 00:11
I have the same problem.  I haven't found a way to import the Rybka 4 openings database that I bought separate.  The instructions don't work.  I may need to return it.
- By ChessMonster (***) Date 2010-12-26 03:23 Edited 2010-12-26 03:28
Where can I find a tutorial regarding the engine personality options of Rybka? In some endgame scenarios, I want to put more weight to keeping some pieces on board rather than exchanging them.

- By e4d6 [de] Date 2011-02-16 19:52

I've a little problem using Deep Rybka 4 64 Bit UCI with chessbase 10.

Starting the engine from a position needs 8 seconds; closing the engine needs 8 seconds again.
On my older computer both steps need only 1 second.

The paramaters are equal with the exception of the hash file.
In my old computer there is a .rsh file given, in my new one it ist <empty> and if I type in a file name it is later empty again.

Does anyone know the solution ?

Thank you and best regards
Previous Next Up Topic Rybka Support & Discussion / Rybka Support / Rybka 4 FAQ (18773 hits)
1 2 3 4 5 Previous Next  

Powered by mwForum 2.23.1 © 1999-2011 Markus Wichitill