Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / BREAKING NEWS 2.3.2!!!! IS OUT!!!!
1 2 Previous Next  
- - By stvs (***) Date 2007-05-17 10:31
testing in playchess server vas says is about 50+elo than 2.3.1!!!!!!
Parent - - By Felix Kling (Gold) Date 2007-05-17 10:50
just a fake.
Parent - - By stvs (***) Date 2007-05-17 10:54 Edited 2007-05-17 10:56
no!!! is true !!! right now on playchess vas testing the new baby!! go to playchess and look!
Parent - - By MidnightBlack Date 2007-05-17 10:57
+50? Lets get serious that would be more than impressive.
Parent - By stvs (***) Date 2007-05-17 10:59
is true!! vas says in one reply is 50 elo better than 2.3.1!! cmon guys go all to playchess!!!!!
Parent - - By Quapsel (****) Date 2007-05-29 11:00
In the rating lists at
http://www.computerschach.de/index.php?option=com_wrapper&Itemid=226
it is described, that 2.3.1 ist 21 (multiple processor) and 49 (single processor) ELO-points weaker than 2.2

Would it realy be such a surprise, if 2.3.2 has the Strength of 2.2 and a bit more (single processor version)?

Quap
Parent - - By Uri Blass (*****) Date 2007-05-29 11:14
You do not have enough games in this list and I guess based on all data that I saw that there is no big difference in playing strength between different rybka's

I see no reason to assume that 2.3.2 will have the strength of 2.2 and a bit more.
My guess is that 2.3.2 will be at least 40 elo better than the next best rybka with the same number of processors in all lists(assuming more than 500 games for both versions).

Uri
Parent - - By Quapsel (****) Date 2007-05-29 12:19
OK the number of games are 860 (rybka 2.2) and 680 (rybka 2.3.1)
and perhaps the error might be 30 ELO-Points

Do you know other Comparisons of the Strengths of 2.2 and 2.3.1, basing on significant more than 500 games?
What do they tell us about the relative strengths?
Can you give us URLs?

Quap
Parent - - By Uri Blass (*****) Date 2007-05-29 12:30
I know 2 different links with both 2.3.1 and 2.2

CCRL 40/4

http://computerchess.org.uk/ccrl/404/rating_list_all.html

Rybka 2.3.1 64-bit 2CPU 3078 +24 −23 80.5% −233.5 23.4% 798  
Rybka 2.2 64-bit 2CPU 3058 +13 −13 74.2% −180.6 28.8% 2513
62.8%

Rybka 2.2 32-bit 2992 +13 −13 74.1% −176.8 28.6% 2266
50.0%
  Rybka 2.3.1 32-bit 2992 +15 −15 74.9% −183.4 26.2% 1805
67.

CEGT 40/20

http://www.husvankempen.de/nunn/40_40%20Rating%20List/40_40%20All%20Versions/rangliste.html

1 Rybka 2.2 x64 2CPU 3015 16 16 1205 72.4 % 2847 35.2 %
4 Rybka 2.3.1 x64 2CPU 3009 20 20 815 73.7 % 2831 36.2 %

11 Rybka 2.3.1 w32 1CPU 2956 25 25 546 71.0 % 2800 33.5 %
13 Rybka 2.2 w32 1CPU 2937 17 17 1184 69.0 % 2799 34.4 %


Uri
Parent - By Quapsel (****) Date 2007-05-29 13:45
Thanx
the first shows Blitz-Results. But you are right: Here I find, that 2.3.1 ist stronger than 2.2 on MP and nearly of equal strengs on SP.

And the second shows nearly eqal strength of these versions 2.2 and 2.3.1, having longer thinking-Time.

OK, my valuation was wrong, it seems.
(sorry, Vas)
And I have to evaluate mmaannyy games to get significant results.

Quap
Parent - - By Felix Kling (Gold) Date 2007-05-17 10:58
Ah, ok, you are right :)
Parent - By stvs (***) Date 2007-05-17 11:21
nice work vas !. lets go rybka fans fire on the forum :)
Parent - - By Steffelino (**) Date 2007-05-17 11:21
Hi!
It is surprising that Vas after a few games dare to say 50 elo better!
I rather wait until an official comment from Vas!
It is a beta version so I guess we have to wait another week.
Parent - - By stvs (***) Date 2007-05-17 11:24
in one reply from vas in playchess, yes indeed he say that!! little more patience and vas tell that here!!
Parent - By stvs (***) Date 2007-05-17 11:31 Edited 2007-05-17 11:38
live report from cnnrybka coming soon: vas announcement the new rybka 2.3.2.(cnnrybka-we are first to news)
Parent - - By Vasik Rajlich (Silver) Date 2007-05-17 12:10
Hi,

I'm just doing some last-minute tests before Leiden. The public release is still not quite ready. It will be soon, but I won't dare to give a day yet.

And yes, in self-play, 2.3.2 clobbers 2.3.1, I've played >1000 blitz games. We'll see how much of this shows up in the official rating lists.

Vas
Parent - - By stvs (***) Date 2007-05-17 12:15
tnx vas. can you confirm the 50 elo diff.in playchess? also is bit more agressive ?
Parent - - By Vasik Rajlich (Silver) Date 2007-05-17 12:21
There are a lot of changes. The most visible ones are related to understanding passed pawns and other pawn majorities, Rybka is better able to distinguish between dangerous ones and harmless ones.

Re. Elo, yes, although this is from blitz self-play. Testers will give a more accurate figure when the version is released.

Vas
Parent - - By Debaser (***) Date 2007-05-17 12:32
Congratulations Vas :)

But I can not imagine why are you still improving so much this 2.x series, instead of just bug fixing. Your engine is already the best!!!

You must be quite sure that you can improve even more in future Rybka versions 3.x, 4.x,...

Two questions about this, perhaps it is very soon to answer, but:

- Will 3.x and the next ones be like 1 and 2 with interim versions, 1.1, 2.2 or just like most of the other engines, for example Fritz 8, Fritz 9, Fritz 10

- On the long term will we enjoy Rybka 8, Rybka 9, Rybka 10...? I mean do you plan to work on chess engines for a long time? Sorry if this is a personal question ;)
Parent - - By plicocf (***) Date 2007-05-17 13:12
Debaser, I don´t understand Vas too. Why improving 2.3.2 so much?
I thinked that Vas was working just in bug fix.
Well, it´s good for us.
Thanks, Vas.

Paulo Soares
Parent - - By Kapaun (****) Date 2007-05-17 13:19
We will see that - depending on how many new bugs he has created while improving the code... ;)
Parent - By Uri Blass (*****) Date 2007-05-17 13:44
I think that he should do some big testing to discover bugs before release and release it only if rybka pass the test.

I suggest the following test:
Take a big file of pgn file and tell rybka to analyze every position on the file for one second.
Do the same with some free source code chess program or with some old rybka that has no known bugs.

If you find a big difference in the evaluation of some position then save them and look at them to see if you have bugs based on analysis of these positions.

Only if after this test you find no bugs release a new version.

Uri
Parent - - By Uri Blass (*****) Date 2007-05-17 13:39
I can explain it.

When I bought rybka2 I hoped for something that is at least 50 elo better than previous version.

Until now rybka2.x failed to provide this hope so vas does everything that the last version will not fail.

From the CEGT we see that even 2 cpu do not give rybka2.x 50 elo improvement relative to rybka1.2

http://www.husvankempen.de/nunn/40_120_ratinglist/ratinglist/rangliste.html

1 Rybka 2.3 64 2CPU 2996 20 20 750 72.1 % 2831 38.1 %
2 Rybka 2.1c 64 2CPU 2994 21 21 750 71.9 % 2831 35.9 %
3 Rybka 1.2f 64-bit 2959 18 18 950 70.3 % 2809 36.8 %

from the 40/20 list we have:

1 Rybka 2.2 x64 2CPU 3013 16 16 1205 72.4 % 2845 35.2 %
5 Rybka 2.1o x64 1CPU 3004 61 57 66 70.5 % 2853 50.0 %
10 Rybka 1.1 x64 2963 17 17 1768 83.1 % 2686 20.4 %
11 Rybka 2.3.1 w32 1CPU 2954 25 25 546 71.0 % 2799 33.5 %

I took the best rybka2.x 2 cpu and the best rybka2.x 1 cpu and the best rybka1.x 1 cpu and again even with 2 cpu we do not have more than 50 elo improvement(note that the rating of the best 1 cpu is probably too big because it has not many games and it is not clear even if 2.3.1 is an improvement relative to 1.1)

only in the blitz rating list you can see significant improvement relative to version 1.x

1 Rybka 2.2 mp x64 2CPU w/Fix 3035 19 19 1100 76.2 % 2833 27.9 %
7 Rybka 2.2 64-bit 2992 24 23 700 73.9 % 2811 28.1 %
10 Rybka 1.2f 64-bit 2945 19 19 1190 75.6 % 2749 23.9 %

Let look at the CCRL list

in the 40/40 list we have the following(again you can see that there is no big difference and the best 2.x is only 6 elo better than the best 1.x)

  Rybka 2.2 64-bit 2999 +35 −34 70.0% −128.8 47.9% 267
61.5%
  Rybka 1.2 64-bit 2993 +21 −20 74.2% −171.1 33.6% 878
54.8

Again when I look at the CCRL at faster time control I find bigger difference.
It seems that version 2.x is designed for blitz and I suspect that 1.2 is simply the best for correspondence time control.

Vasik needs to give something to the correspondence players who use a single processor to justify their decision to buy rybka.
Maybe he is afraid that bug fix will not be enough to produce at least 50 elo improvement relative to previous version so he decided to do more improvements.

Uri
Parent - - By Debaser (***) Date 2007-05-17 14:18
Hi Uri. I use Rybkaw32 1CPU ;)

CETG 40/20 +21

11  Rybka 2.3.1 w32  2954  25  25  546  71.0 %  2799  33.5 %
14  Rybka 1.2f w32  2933  25  25  605  73.1 %  2760  27.4 %

CETG 40/4 +50

11  Rybka 2.2 32-bit  2938  16  16  1345  69.9 %  2791  30.9 %
17  Rybka 1.2f 32-bit  2888  17  17  1245  68.5 %  2753  27.2 %

CCRL 40/4 +33

Rybka 2.3.1 32-bit  2995  +15  −15  75.4%  −187.4  25.9%  1751
Rybka 1.2f 32-bit  2962  +14  −14  67.2%  −123.1  33.1%  1935

CCRL 40/40 +14

Rybka 2.2 32-bit  2991  +19  −19  71.4%  −146.7  36.7%  949
Rybka 1.2 32-bit  2977  +19  −18  70.7%  −149.4  34.9%  1075

CSS 10/10 +54. 2.3 was a spep back. I miss this rating list :(

4  Rybka 2.2 32-bit  2980  20  45                520  360  112  48  416,0  21,5%  80,0%  2739 
6  Rybka 1.2 32-bit  2926  18  34                666  427  157  82  505,5  23,6%  75,9%  2727 

So there is an improvement here.
Parent - - By Uri Blass (*****) Date 2007-05-17 15:51
The improvement seems to be only for fast time control.

You can see the tendency for the 32 bits.

40/4 +50 or +33
40/20 +21
40/40 +14

It is logical to believe that at correspondence time control it may be negative improvement.

Note that CSS 10+10 is faster time control than 40/20 or 40/40

Uri
Parent - By Debaser (***) Date 2007-05-17 16:41
Probably you are right, we only have left:

CETG 40/120 +37

1  Rybka 2.3 64 2CPU  2996  20  20  750  72.1 %  2831  38.1 %
3  Rybka 1.2f 64-bit  2959  18  18  950  70.3 %  2809  36.8 %

But of course this is with 2 CPUs
Parent - - By Gaмßito (****) Date 2007-05-17 16:31
When I bought rybka2 I hoped for something that is at least 50 elo better than previous version.

No,
You are just forgetting the result by SSDF where they demonstrated a good improvement of 53 points under slow time controls, between 2.3.1 and 1.2 and using only 1 CPU. Did not you believe in the improvement?

1. Rybka 2.3.1 256MB Athlon 1200 MHz  2962
2. Rybka 1.2   256MB Athlon 1200 MHz  2909

Regards,
Gambitto.
Parent - By Uri Blass (*****) Date 2007-05-17 17:04
result of ssdf is dependent on book and 2.3.1 used better book than 1.2
I do not care about book but about the engine.

Uri
Parent - By albitex (***) Date 2007-05-28 23:17
Uri, I have played in March / April, on playchess.com against a "Rybka 2.4."
This Rybka was a strange name, had been enrolling for that same day, and then disappeared.
I think that Vas already has from some something in the yard.
But you because not you think about personally modifying the program?
Anybody that I/you/he/she am braco with the C++?
Parent - - By Vasik Rajlich (Silver) Date 2007-05-19 07:52
I'm not sure yet about the version system for Rybka 3 and later. We'll cut back a bit on the release frequency, but it's not clear how much.

I feel like right now, some of the improvements are 'must-have'. Maybe later, slower releases will make more sense.

Re. long term, you never know. I certainly have no plans to disappear.

Vas
Parent - - By jps7 (**) Date 2007-05-25 04:59
Hi Vas,

I think too many versions of Rybka release are good as it keeps Rybka on top and and in news as well. Cutting the release frequency may enable rivals to take over and will also make it boring to play with same engine for years. What do you think. Being an MBA, I feel frequent releases keeps customers intact with you. I only feel that having own Rybka GUI can help Rybka users a lot.

JP Singh
Parent - By Vasik Rajlich (Silver) Date 2007-05-25 07:52
There are pros and cons. The pros is that customers get improvements into their hands faster. The cons is the logistics issue.

Vas
Parent - - By Kapaun (****) Date 2007-05-17 12:34
It will be soon, but I won't dare to give a day yet.



Depending on the results of Leiden? ;-)
Guys, let's all cross our fingers that Vas is going to trash all his competitors! :-D
Parent - - By -valdimor- (**) Date 2007-05-17 13:19
although i'm really excited about this new version,but i still say Rybka 2.1d3 mp forever :)

Vas you already have the perfect engine!
Parent - By Kapaun (****) Date 2007-05-17 13:33
Nah - I don't think so. 2.1d3 may be pretty good, but it is not the best one...
Parent - - By exigentsky (***) Date 2007-05-17 15:24
Congratulations Vas! We all wanted Rybka 2.3.2 to be as stable and bug free as possible because it would be the version for which the entire 2.x series would be judged for a long time, but I'm sure none of us expected such improvement in strength! Now, I think it's pretty much guaranteed that no one will catch up to Rybka while 3.0 is developed.

Thanks a lot!
Parent - By RFK (Gold) Date 2007-05-17 18:04
Absolutely! It is a great step forward and I look toward Rybka 2.3.2's release.

sidserious
Parent - - By Nelson Hernandez (Gold) Date 2007-05-17 18:38
No day given, that's what I want to hear!  In fact, just avoid a "release" entirely and hide it instead. 

Create a "Hunt for Rybka 2.3.2 Adventure" where we would need to piece together clues, solve riddles and mysteries, perform mathematical feats, engage in cryptanalysis, and finally find a buried box somewhere in the world with GPS positioning and a metal-detector or Geiger counter.  It would be an even bigger sensation than Freestyle and the usual suspects would all be vying with each other to get to the update first!  (However it would be very, very important that this release solve the bishop underpromotion issue.)
Parent - By RFK (Gold) Date 2007-05-17 20:54 Edited 2007-05-17 20:58
Nelson,

It sounds like all the work on Rybka 2.3.2 is done. So! Well! I guess you'll just have to deal with the posibility of an unresolved "bishop underpromotion issue."

In point of fact, there really is no need for anymore attempts to posts perceived real or imagined issues regarding the "old" if not "historic" Rybka 2.3.1. 

Oh? But, wait ! What is this? Looks like an email from, Sasha!

[Dear Robert,

....New version Vasik promises to finish at the end of the May...]

regards
sidserious-

P.S. Has anyone heard from Jessica as to when she is going to have a family reunion?
Parent - - By Gaмßito (****) Date 2007-05-18 04:54 Edited 2007-05-18 05:23
Vas,

How it finished your 1000 blitz test and what specific time control used?

By the way, really many thanks for this new version, continue with your excellent work and my best desires for you in the Leiden Tournament.

Regards,
Gambitto.
Parent - By Vasik Rajlich (Silver) Date 2007-05-19 07:56
Actually, there are four sets of tests, still ongoing. The results dropped a bit now, to around +40 Elo. We're at around 2000 games now. Also, some tests of the MP version give a significantly lower figure, but with fewer games. All of this will be straightened out in the next few weeks.

Vas
Parent - By Quapsel (****) Date 2007-05-29 11:06
Hi Vas,

what do you think, how is the strength of 2.3.2 compared to 2.2?
Would 'clobber' be here a fitting verb too?
(AFAIK 2.3.1 is described as significant weeker than 2.2 at several places)

Quap
Parent - - By lkaufman (*****) Date 2007-05-17 15:51
     I would like to comment a bit on where this nearly fifty point increase (if it holds up) came from, since I've been working on the evaluation portion during this period. Some of it came from bug fixes and improved search by Vas. A good deal of it came from what I would call "chess bug" fixes. By this I mean values in the evaluation function (put there by an automated procedure) that were (to my thinking at least) obviously wrong, i.e. "chess bugs". Just to illustrate, Rybka considered that in the endgame a bishop on d2 was generally a bit better than on b2 because it is "closer to the center". But of course this is silly; they have equal mobility on an empty board, but on b2 it points towards the center, whereas on d2 it points towards the edge. Fixing this "chess bug" was probably worth less than 1 Elo point, but there were lots of similar little "chess bugs", and it seems that every time I fixed this type of bug we gained a point or so, and these points add up. Finally, roughly a third of the gain came from a simple but important piece of chess knowledge that was missing from Rybka, which I won't name as we don't want to help the competition!
     I would just like to add that there is no shortage of things to improve for future versions!
Parent - - By Debaser (***) Date 2007-05-17 16:45
Thanks for clarifying this, we forget that Rybka team is not only Vas ;)

I bet that this "important piece of chess knowledge" is not bishops underpromotions?
Parent - By George Tsavdaris (****) Date 2007-05-17 17:43

>I bet that this "important piece of chess knowledge" is not bishops underpromotions?


It's not. Bishop underpromotions fix would give less than 1 ELO point, so that can't be the 1/3 of the gain....
Parent - - By boo! (**) Date 2007-05-17 17:41 Edited 2007-05-17 17:50
The improvement in the search wouldn't happen to be a DTS(Dynamic Tree Splitting) implementation? :)
Parent - - By Vasik Rajlich (Silver) Date 2007-05-19 07:59
There are no changes to the multi-processor implementation yet. Or at least, there are no purposeful changes.

Vas
Parent - - By cma6 (****) Date 2007-05-30 01:31 Edited 2007-05-30 01:43
There are no changes to the multi-processor implementation yet. Or at least, there are no purposeful changes.
Vas

  Does this mean that 2.3.2 mp is not expected to be any stronger than 2.3.1 mp?!  Vas, say it ain't so!

  In any case, a real increase of as much as 35 ELO  points measured at very slow play (correspondence) for 2.3.2 over 2.3.1 would be an amazing, unexpected achievement, for you and LK. I only hope that the mp version will also be improved.
Parent - By Uri Blass (*****) Date 2007-05-30 06:17
No

No change to the multi-processor implementation means that rybka will not earn more speed from more processors.

I guess 2.3.2 can be expected to be significantly stronger than 2.3.1 and I disagree that a real increase of 35 elo at slow play would be amazing unexpected achievement.

Rybka is full of bugs and only fixing the bugs may give her more than 35 elo

Uri
Parent - - By Uri Blass (*****) Date 2007-05-30 06:23
No

No change to the multi-processor implementation means that rybka will not earn more speed from more processors.

I guess 2.3.2 can be expected to be significantly stronger than 2.3.1 and I disagree that a real increase of 35 elo at slow play would be amazing unexpected achievement.

Rybka is full of bugs and only fixing the bugs may give her more than 35 elo at slow time control.

Based on my memory one of the bugs is hash bugs that can cause her to have short pv because of deleting important information from the hash.

This bug is not important in blitz but clearly important at slow time control so maybe only fixing bugs(including the bugs that cause her to have mate score when there is no mate that is important in all time controls) can give rybka only 20 elo at blitz but I believe that it can give rybka something near 50 elo at longer time control.

When we know that rybka2.3.2 has more knowledge relative to 2.3.1 35 elo improvement relative to 2.3.1 will be a very big disappointment.

Uri
Up Topic Rybka Support & Discussion / Rybka Discussion / BREAKING NEWS 2.3.2!!!! IS OUT!!!!
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill