Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Support / Rybka 4.1 Release Notes
1 2 3 Previous Next  
- - By Vasik Rajlich (Silver) [pl] Date 2011-03-05 07:17 Edited 2011-03-08 06:28
Overview

Rybka 4.1 is the latest UCI Rybka chess engine and is a free update for Rybka 4 customers. This is a relatively minor version - a number of bugs which existed in Rybka 4 are fixed, but there are no search or evaluation changes of any kind. The engine parameters are also mostly unchanged.

Update Instructions

In order to obtain Rybka 4.1, you will need to have the original Rybka 4 executables, and you will need to download the attached Rybka41Update.exe application. Then:

- Save Rybka41Update.exe to your hard drive and run it (usually by double-clicking the icon) and follow the instructions in the updater.
- The updater will look for existing Rybka programs in the directory you supply it with and create Rybka 4.1 versions alongside any Rybka 4 programs that are found.
- You can add these new Rybka 4.1 programs to your chess GUI of choice.
- After the update process the Rybka41Update.exe program can be deleted, but you may want to keep it for backup purposes.

The Rybka 4 to 4.1 updater works for both Deep Rybka 4 and Rybka 4.

Note 1: The updater should be pointed to the directory in which the Rybka 4 executables reside. If you have trouble finding the Rybka 4 executables, you can perform a system file search for "Rybka*.exe".
Note 2: The updater uses JoJoDiff by Joris Heirbaut, it is not necessary to download it, but the source of JoJoDiff can be obtained from http://jojodiff.sourceforge.net/

Thanks To

A big thanks to Nick Carlin for putting together the updater application, as well as to the Rybka team and the Beta testers for their help.
Attachment: Rybka41Update.exe (1855k)
Parent - - By Vasik Rajlich (Silver) [pl] Date 2011-03-06 06:27
Addendum: Rybka 4.1 also has two unofficial and unsupported features. More info can be found here.

Vas
Parent - By Dr.X (Gold) Date 2011-03-07 10:49
Thanks for coming through!
Parent - - By irulats (****) [ie] Date 2011-03-07 18:18
Fantastic! Thanks for making the effort. Hope all's well with you and yours.
Parent - - By Vasik Rajlich (Silver) [pl] Date 2011-03-07 18:33
Thanks - it is. :smile:

Vas
Parent - - By keoki010 (Silver) [us] Date 2011-03-08 00:09
Thanks for the update.  Never mind all the idiocy!:sad:
Parent - - By cma6 (****) Date 2011-03-09 04:16
Is there any reason to update to Rybka 4.1 since the strength of the engine is the same as ver. 4.0?
                                       Thanks, CMA
Parent - - By Uly (Gold) [mx] Date 2011-03-09 05:44
If all you use it for is engine engine matches, then probably not.
Parent - By ernest (****) [fr] Date 2011-03-09 13:44

> engine engine matches, then probably not


How about the LTC fantasm? :grin:
Parent - - By cma6 (****) Date 2011-03-10 01:17
Uly:
I use Aquarium/Rybka for correspondence analysis. Does 4.1 offer any enhancements over 4.0 for the correspondence player?
                             Thanks, CMA
Parent - - By Uly (Gold) [mx] Date 2011-03-10 02:18
Yes.

One clear concept here:

Due to lack of Persistent Hash and radical change in playing style (low redundancy), Rybka 3 and Rybka 3 Dynamic are still in my core of main engines, are still useful, and have NOT been obsoleted even at this point. I'll continue to use them for these reasons.

On the other hand, Rybka 4.1 clearly obsoletes Rybka 4, there's absolutely* no reason to keep using Rybka 4, and there's no reason for not upgrading. Gone bugs like the stall one beat laziness.

Just switch on "Display Upperbounds" and "Preserve Analysis", and you'll notice the improvement is not just placebo.

(*) Never say "absolutely", probably if you use mindbraker's settings for analysis you'd want to keep R4 to use them, wait till equivalent settings appear for R4.1, or make them yourself.

Disclaimer: I'm still using Rybka 4.1 RC2 due to upgrade laziness, but I'll upgrade as soon as I'm tired of seeing the White King millipawns parameter. Oh, the irony.
Parent - - By Dragon Mist (****) [hr] Date 2011-03-10 06:34

> Just switch on "Display Upperbounds" and "Preserve Analysis", and you'll notice the improvement is not just placebo.


Hold on right there! You mean there is no more exploding of branching factor when fail low occurs, and those 2 are active?? Would you recommend it be on when doing infinite? And why? :-)
Parent - - By Uly (Gold) [mx] Date 2011-03-10 21:58

> You mean there is no more exploding of branching factor when fail low occurs


No, eh, there's a bug that causes Rybka sometimes not to "Always Score Main Move", in such cases she will take the long road of evaluation and won't show the exact score of the move, so you get the worst of both worlds! I consider the setting broken and not recommend to use it.

And, anyway, I recommend the user to take interactive action when he sees a fail low (Vempele recommended for the Rybka 2 series to switch to MultiPV when one sees a fail low, that's the workaround for Rybka 3 Persistent Hash, but what I've been doing is just forcing the failing move, which turns it into a fail high, and then Rybka will reach relative depth +1, or even +2 much faster) which requires "Display upperbounds" ON to see them.

"Preserve Analysis" is recommended if you interact with the position, Rybka is doing a great job remembering old variations analyzed, and their transpositions.
Parent - - By Vempele (Silver) [fi] Date 2011-03-10 22:16

> No, eh, there's a bug that causes Rybka sometimes not to "Always Score Main Move"


I thought Vas fixed that? You found the bug pretty early and he never mentioned not fixing it...
Parent - By Uly (Gold) [mx] Date 2011-03-11 00:18
I take all the bugs in where he didn't say if he fixed them as "not fixed".

No idea if the bug is fixed or not, my point is that, even if it is, sitting to wait for a fail low to be resolved is not a good idea, having the parameter OFF is good for real fail lows where the main move does change. One would need to wait for the fail lows to no longer explode for the parameter to be useful, as it is, knowing the score of the main move can be done faster manually.
Parent - - By cma6 (****) Date 2011-03-11 01:24 Edited 2011-03-11 01:33
Uly:
Thanks for the recommendation to upgrade. However, some of your comments were not clear.
"And, anyway, I recommend the user to take interactive action when he sees a fail low (Vempele recommended for the Rybka 2 series to switch to MultiPV when one sees a fail low, that's the workaround for Rybka 3 Persistent Hash, but what I've been doing is just forcing the failing move, which turns it into a fail high, and then Rybka will reach relative depth +1, or even +2 much faster) which requires "Display upperbounds" ON to see them.
Preserve Analysis is recommended if you interact with the position, Rybka is doing a great job remembering old variations analyzed, and their transpositions."

1) Since one has to interact in Infinite Analysis (why else would you use I.A.?) and one is supposed to interact in IDeA, are you saying we should use Preserve Analysis? Thr experts said that we should not use Preserve Analysis in R4. What has changed? I haven't seen a recommendation on Preserve Analysis from Vas.

2) What do you mean by "taking action" when one see a fail low? How does one know if one is seeing a fail low or fail high? And what action are you recommending?

3) Here were the recommended engine settings by the experts for R4 (including piece weightings from LK):
  Use the following "Display Upper Bounds"; "Always score main move"; "Use Large Pages";
Do not use "Preseve Analysis"; Max CPU = 4 (if applicable); Nalimov Caches = 128.
  Ns and Bs = 12; R = 6; Q & P = 0;

Do you suggest any changes (including Preserve Analysis) from the recommended settings in 3) for ver. 4.1?
Also, ver. 4.1 has a new setting: "Bs are opposite colored penalty". Any change to default of "0"?

4) In ver. 4.0, one could review the Log file to see if Large Pages were used. If so, one would see in the Log file, something like this: "Using Large Pages, size = ...."

However, in the Log file for ver. 4.1, I do not see that. So how does one know if Large Pages are being used?
All one sees in the new Log file is this: "Large Page Use enabled (had been disabled)" without any size.

                           Thanks for the clarification,
                                                              CMA
Parent - - By Uly (Gold) [mx] Date 2011-03-11 02:27

> Thr experts said that we should not use Preserve Analysis in R4. What has changed?


It was tested (really! Rybka 3 "preserve analysis" was really bad, and recommended to be off. But Rybka 4 "preserve analysis" is great. I didn't get to test it until Rybka 4.1. If someone told you to put it off for Rybka 4, it's probably because he didn't test it and assumed it was like R3's PA).

It should be off for games, or for forward analysis when positions are never revisited. ON for when you're going to revisit positions and transpositions. I never tested IDEA with it, should help in theory.

> 2) What do you mean by "taking action"


Switching to MultiPV=2 when a fail low is hit (if you want an alternative move ASAP), or forcing the move that is failing (if you want to know its score ASAP).

>How does one know if one is seeing a fail low or fail high?


After having "Display upperbounds" ON, Shredder Classic shows ++ besides a score in fail highs, and -- in fail lows. Other interfaces should show something similar.

>Do you suggest any changes (including Preserve Analysis) from the recommended settings in 3) for ver. 4.1?


I don't know from where those settings come from, I'll only comment on them:

>  Use the following "Display Upper Bounds"


Yes.

>"Always score main move"


No (unless you're not going to "take action".)

>"Use Large Pages"


Yes (if you know what you're doing).

>Do not use "Preseve Analysis"


On what basis? If the idea is "it was bad for Rybka 3, must be bad for Rybka 4", then I'd say, use it.

>Max CPU = 4 (if applicable)


Set them to 2048 so Rybka uses the max you have (e.g. this is a bad one if you got a 12core.)

> Nalimov Caches = 128.


? (I don't use tablebases, probably yes)

>  Ns and Bs = 12; R = 6; Q & P = 0;


No (that's I guess for analysis of human games? For analysis of correspondence games I'd recommend Default, though, anyway, we've moved from centipawns to millipanws, those values are no longer valid)

>4)


I don't use Large pages, so I don't know.
Parent - - By cma6 (****) Date 2011-03-11 02:49
"Preserve analysis should be off for games, or for forward analysis when positions are never revisited. ON for when you're going to revisit positions and transpositions. I never tested IDEA with it, should help in theory."

If one is analyzing one's correspondence positions, where there is much intervention, should PA be on or off. I could not tell from the above sentence. It is not practical to constantly change an engine parameter like PA.

"After having "Display upperbounds" ON, Shredder Classic shows ++ besides a score in fail highs, and -- in fail lows. Other interfaces should show something similar." 
Aquarium does not show anything for a Fail High or Fail Low, although I do not know what they are.

"Always score main move"
No (unless you're not going to "take action".)

Did you mean that one should use "Always Score Main Move" if one is goint to "take action" or if one is not going to take action?  With all the negatives, I could not figure out the meaning of your sentence. ("No", "unless" and "not" = 3 negatives.)
The experts told us to definitely use "Always Score Main Move" for R3 and R4.
Parent - By Uly (Gold) [mx] Date 2011-03-11 05:08

> If one is analyzing one's correspondence positions, where there is much intervention, should PA be on or off.


On.

>Aquarium does not show anything for a Fail High or Fail Low, although I do not know what they are.


Do you get the same behavior when having "Display upperbounds" on and off? If so, it seem like an Aquarium bug.

> Did you mean that one should use "Always Score Main Move" if one is goint to "take action" or if one is not going to take action?


If:

You're going to not do anything in a fail low: set Always Score Main Move ON

If:

You're going to interact with the engine when you notice a fail low: set Always Score Main Move OFF

> The experts told us to definitely use "Always Score Main Move" for R3 and R4.


When in doubt, consult back those experts. I'd like to have a talk with them, specially if they didn't test R4's PA=ON at all.
Parent - - By Vempele (Silver) [fi] Date 2011-03-11 06:43 Edited 2011-03-11 06:53

> The experts told us to definitely use "Always Score Main Move" for R3 and R4.


The expert who told you this was mistaken - there is no "Always Score Main Move" in R3.

> Aquarium does not show anything for a Fail High or Fail Low


It shows > for a white fail high or a black fail low and < for a black fail high or a white fail low. Technically, it should be  ≥ and ≤.

> although I do not know what they are.


How could you possibly know what Aquarium shows or doesn't show for them?
Parent - By Uly (Gold) [mx] Date 2011-03-11 07:08

> It shows > for a white fail high or a black fail low and < for a black fail high or a white fail low. Technically, it should be  ≥ and ≤.


Thanks. I'll add that, when Rybka (at default settings) goes on to search for alternative moves without giving a score to the main move, it's a fail low.
Parent - - By cma6 (****) Date 2011-03-11 18:23
Thanks to Uly for the very useful clarification and to Vempele for the information about Fail High/Fail Low.
   Vempele, where in the GUI do I look for > or < ?
Parent - - By Vempele (Silver) [fi] Date 2011-03-11 18:27
To the left of the score. Do the symbols maybe remind you of something you learned in first grade math? :razz:
Parent - - By cma6 (****) Date 2011-03-11 21:09
Vempele, thanks.
1st grade is so long ago for me, it could have been Jurassic era.
Parent - By keoki010 (Silver) [us] Date 2011-03-11 21:25
For a lot of us... :eek::lol:
Parent - - By cma6 (****) Date 2011-03-12 03:25 Edited 2011-03-12 21:27
With White to move, does > mean a "White Fail high"; and with Black to move, it means "Black Fail Low"?

   What I frequently notice before the score with White to move is "[> (+1.08)] d=21 33...Be8, etc."

Then after 20 seconds or so, the ">" will disappear. 

However, sometimes the ">" never disappears. I suppose that means a "White Fail High"?

The odd thing about Fail High is the for the next PV, it may not end up higher than +1.08 but at +1.03
Parent - - By Vempele (Silver) [fi] Date 2011-03-12 08:36

> With White to move, does > mean a "White Fail high"; and with Black to move, it means "Black Fail Low"?


Yes.

> Then after 20 seconds or so, the ">" will disappear.


Disappear? I hope you just mean the next PV doesn't have the ">" (i.e. the score is exact, not a lower bound), because otherwise it'd be a really weird bug.
Parent - - By cma6 (****) Date 2011-03-12 16:53
Vempele, I did mean that the next PV no longer has the ">". So a "lower bound" is synonymous with a "Fail High"?
Parent - - By Vempele (Silver) [fi] Date 2011-03-12 17:04
K is a lower bound of x: K ≤ x. Fail high: The search result is above the window so the returned score is a lower bound of the true score.
Parent - By cma6 (****) Date 2011-03-12 20:31
Vempele,
Thanks for the tutorial "Fail Highs in one easy lesson."
               CMA
Parent - By cma6 (****) Date 2011-03-14 03:44 Edited 2011-03-14 03:47
I have been working with infinite analysis in one of my correspondence games over the last couple of days. I can say that Preserve Analysis really works as it was supposed to in earlier versions.

If R4.1 shows a score of -0.07 after 23 Rd1 with d=25, then I can key back (gradually) during i.a. to 22 Kf1; and R41 will eventually realize that the correct score is -9.07 after 22 Kf1 and not -0.01 as originally scored after 22 Kf1. Prior to this implementation of Preserve Analysis, it was very difficult to force the correct retrospective scores with infinite analysis.
Parent - - By kblaine (**) Date 2011-03-08 14:25
Does 4.1 fix the problem where Rybka hangs in mv mode? You know, sits on one depth for hours and refuses to advance even at a relatively low depth such as 15.
Parent - By Vempele (Silver) [fi] Date 2011-03-08 14:33 Edited 2011-03-08 14:35
Yes. It also fixes some other stalls.

OT: What happened to my location?
Parent - By titanium cranium (***) [us] Date 2011-03-17 22:07
Her assessments with this version seem to be more erratic or vary more from depth to depth than with 4.0. Just an observation, not scientific.
Parent - By Indrajit (***) [in] Date 2011-03-07 10:37
Awesome..... :lol:
Parent - By ppipper (*****) [es] Date 2011-03-07 11:39
I never thought this was going to happen. Now it is time to say sorry, just in case.

Thanks!
Parent - - By Bouddha (****) [ch] Date 2011-03-07 17:05
I do not even find where my Deep Rybka 4 is installed.

Under ChessBase/Fritz GUI where does it install that stuff ?

regards
Parent - - By Dragon Mist (****) [hr] Date 2011-03-07 17:09
Try something like Program files - Common files - Chessbase - Engines.uci - Rybka 4 or something like that.
Parent - By Bouddha (****) [ch] Date 2011-03-07 20:29
Thank you !
Parent - - By Vempele (Silver) [fi] Date 2011-03-07 17:20
Run the engine, find it in Task Manager, Right-click->Open File Location.
Parent - By Bouddha (****) [ch] Date 2011-03-07 20:22
Thank you !
Parent - - By Vasik Rajlich (Silver) [pl] Date 2011-03-08 06:29
This doesn't work for me (although it's a cool idea) - there is no 'Open File Location' item in the context menu.

Vas
Parent - - By Vempele (Silver) [fi] Date 2011-03-08 07:23
Do you at least have 'Properties'? Copy/paste the location into Explorer.
Parent - By Vasik Rajlich (Silver) [pl] Date 2011-03-08 18:01
Ok, I see what the issue is: this feature is supported under Windows 7 x64 but not Windows XP x64.

Vas
Parent - - By Dragon Mist (****) [hr] Date 2011-03-08 16:15
Even cooler: run the GUI (if in Fritz), then Create UCI engine, then browse to see where is R4 located. :wink:
Parent - - By Vasik Rajlich (Silver) [pl] Date 2011-03-08 18:03
If it works .. :smile:

Vas
Parent - By Radost (***) [de] Date 2011-03-08 20:04
Today I got the info about a new Update (No 20) for Fritz12-GUI and also I found a special page about "Upgrade your Rybka 4 / Deep Rybka 4 engine" (see http://www.chessbase.com/products/rybka4/index.htm).

Greetings, Frank
Parent - By Gaмßito (****) [cr] Date 2011-03-10 07:38
Vas,

Thanks a lot! I really can't believe it. You finally made a fix!!! :smile:

Best Regards,
Gaмßito.
Parent - - By VolMike (*) [ru] Date 2011-03-07 19:24
Bishop underpromotion feature is still not added...
Parent - By vroger007 (**) [be] Date 2011-03-29 19:32
+1
Up Topic Rybka Support & Discussion / Rybka Support / Rybka 4.1 Release Notes
1 2 3 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill