SWCR-64: Rybka 4 Exp. 61 x64 scheduled for Dec. 27th, 2010

Discussion of computer chess matches and engine tournaments.

Moderator: Ras

Frank Quisinsky
Posts: 6936
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: After 465 of 1.000 ... download games, settings!

Post by Frank Quisinsky »

Hi there,

I got different questions to this round robin Rybka 4 x64 Exp. 61 test.

Here what I have ...

Database with 465 games so far, included are the Shredder Classic GUI comments (ply, time, eval), bayesian file, ELOstat "programs" file and the settings I used.

http://www.amateurschach.de/download/sw ... 61-465.zip

= 0.8Mb

Best
Frank
Frank Quisinsky
Posts: 6936
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: After 514 of 1.000 ...

Post by Frank Quisinsky »

Code: Select all

   1 Houdini 1.5 x64              3005   22   21  1021   82%  2747   26% NEW +  56
   2 Rybka 4 x64 Exp. 61          2979   30   29   514   80%  2745   25% NEW +  39 to Rybka 4 default
   - Rybka 4 x64 Exp. 42          2960   23   22   900   78%  2730   26% NEW +  20 to Rybka 4 default
   - Houdini 1.03a x64            2949   22   21   920   79%  2719   29% 
   - Rybka 4 x64                  2940   17   17  1520   80%  2700   29% 
   3 IvanHoe B49jA x64            2933   19   18  1141   76%  2736   33% NEW +  10
Denzwell
Posts: 40
Joined: Fri Mar 05, 2010 2:22 pm

Re: After 400 of 1.000 ...

Post by Denzwell »

gerold wrote:
Denzwell wrote:This Night i will try Rybka 4 Exp. 61 against H1.5 , IvanHoe B49jx64 and stockfish 1.9.1.

But i think H1.5 will win ....
What TC and books will you use Max.

Be looking forward for the results.

Best,
Gerold.
Hi

I use Perfect book 2010. 2 CPU for each engine. Timer 7 minutes.
Chessbase GUI

Live : http://www.afromaniak.com/lives/index.htm
gerold
Posts: 10121
Joined: Thu Mar 09, 2006 12:57 am
Location: van buren,missouri

Re: After 400 of 1.000 ...

Post by gerold »

Denzwell wrote:
gerold wrote:
Denzwell wrote:This Night i will try Rybka 4 Exp. 61 against H1.5 , IvanHoe B49jx64 and stockfish 1.9.1.

But i think H1.5 will win ....
What TC and books will you use Max.

Be looking forward for the results.

Best,
Gerold.
Hi

I use Perfect book 2010. 2 CPU for each engine. Timer 7 minutes.
Chessbase GUI

Live : http://www.afromaniak.com/lives/index.htm
Thanks Max.
Interesting results.

Best,
Gerold.

P.S. Will you be running the test over with different setting for
R4 with the same engines.
gerold
Posts: 10121
Joined: Thu Mar 09, 2006 12:57 am
Location: van buren,missouri

Re: After 514 of 1.000 ...

Post by gerold »

Hi Frank.
In the other forum someone is testing about 75 R4 engines with
different exp. setting vs. around 25 other engines.
My question is playing so many R4 engines vs 25 other engines and
trying to get the strongest setting for R4 . Do you think this is the
way to go. One R4 setting is suppose to be 50 elo stronger than
the regulator setting for R4. No H 1.5 in his tests.

Best,
Gerold.
User avatar
Houdini
Posts: 1471
Joined: Tue Mar 16, 2010 12:00 am

Re: After 514 of 1.000 ...

Post by Houdini »

gerold wrote:Hi Frank.
In the other forum someone is testing about 75 R4 engines with
different exp. setting vs. around 25 other engines.
My question is playing so many R4 engines vs 25 other engines and
trying to get the strongest setting for R4 . Do you think this is the
way to go. One R4 setting is suppose to be 50 elo stronger than
the regulator setting for R4. No H 1.5 in his tests.

Best,
Gerold.
It's probably important to make a distinction between "objective strength" and "higher rating".

Most of the "Exp XX" contempt settings do not make Rybka 4 objectively stronger, they make it avoid draws in order to score higher against weaker opponents. If the rating list contains a lot of weaker engines the Exp settings may produce a higher rating.

But objectively speaking contempt settings do not improve the "strength" of a well-tuned engine like Rybka 4, it should score slightly less against its peers.

Robert
gerold
Posts: 10121
Joined: Thu Mar 09, 2006 12:57 am
Location: van buren,missouri

Re: After 514 of 1.000 ...

Post by gerold »

Houdini wrote:
gerold wrote:Hi Frank.
In the other forum someone is testing about 75 R4 engines with
different exp. setting vs. around 25 other engines.
My question is playing so many R4 engines vs 25 other engines and
trying to get the strongest setting for R4 . Do you think this is the
way to go. One R4 setting is suppose to be 50 elo stronger than
the regulator setting for R4. No H 1.5 in his tests.

Best,
Gerold.
It's probably important to make a distinction between "objective strength" and "higher rating".

Most of the "Exp XX" contempt settings do not make Rybka 4 objectively stronger, they make it avoid draws in order to score higher against weaker opponents. If the rating list contains a lot of weaker engines the Exp settings may produce a higher rating.

But objectively speaking contempt settings do not improve the "strength" of a well-tuned engine like Rybka 4, it should score slightly less against its peers.

Robert
Thanks for your answers Robert.
Yes i can see what you mean on the rating of the engine.

Best,
Gerold.
beram
Posts: 1187
Joined: Wed Jan 06, 2010 3:11 pm

Re: After 514 of 1.000 ...

Post by beram »

Houdini wrote:
gerold wrote:Hi Frank.
In the other forum someone is testing about 75 R4 engines with
different exp. setting vs. around 25 other engines.
My question is playing so many R4 engines vs 25 other engines and
trying to get the strongest setting for R4 . Do you think this is the
way to go. One R4 setting is suppose to be 50 elo stronger than
the regulator setting for R4. No H 1.5 in his tests.

Best,
Gerold.
It's probably important to make a distinction between "objective strength" and "higher rating".
..

But objectively speaking contempt settings do not improve the "strength" of a well-tuned engine like Rybka 4, it should score slightly less against its peers.

Robert
That is precisely the point :-)
Frank Quisinsky
Posts: 6936
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: After 647 of 1.000 ...

Post by Frank Quisinsky »

Code: Select all

Rank Name                          Elo    +    - games score oppo. draws 
   1 Houdini 1.5 x64              3005   22   21  1026   82%  2748   26% NEW +  56
   2 Rybka 4 x64 Exp. 61          2967   26   26   647   79%  2744   26% NEW +  27 to Rybka 4 default
   - Rybka 4 x64 Exp. 42          2960   23   22   900   78%  2730   26% NEW +  20 to Rybka 4 default
   - Houdini 1.03a x64            2949   22   21   920   79%  2719   29% 
   - Rybka 4 x64                  2940   17   17  1520   80%  2701   29% 
   3 IvanHoe B49jA x64            2933   19   18  1146   76%  2737   33% NEW +  10
Frank Quisinsky
Posts: 6936
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: After 831 of 1.000 ...

Post by Frank Quisinsky »

Code: Select all

Rank Name                          Elo    +    - games score oppo. draws 
   1 Houdini 1.5 x64              3004   22   21  1034   82%  2748   26% NEW +  56
   2 Rybka 4 x64 Exp. 61          2963   23   22   831   78%  2744   26% NEW +  24 to Rybka 4 default
   - Rybka 4 x64 Exp. 42          2959   23   22   900   78%  2729   26% NEW +  20 to Rybka 4 default
   - Houdini 1.03a x64            2948   22   21   920   79%  2718   29% 
   - Rybka 4 x64                  2939   17   17  1520   80%  2700   29% 
   3 IvanHoe B49jA x64            2932   19   18  1154   76%  2737   33% NEW +  10
Best
Frank