Bright Spark

[Ovyron]

I made this tourney to test the new Bright.



Settings
Athlon 2.1Ghz 32bit
ChessPartner GUI
1'+ 7" with ponder off
64MB hash each engine
No tablebases
4 Rounds Gauntlet, 32 games
No books for first two rounds, forced one move for other two rounds

Midway results:

Code: Select all

Bright Spark
2008.03.29 - 2008.03.30
                             Score       br Gl Ry Sp Th Fr br To Na
------------------------------------------------------------------
 1: bright-0.3a              10.0 / 16   XX 00 =0 01 10 =1 11 11 11
 2: Glaurung 2.0.1            2.0 /  2   11 XX .. .. .. .. .. .. ..
 3: Rybka 1.0 Beta 32-bit     1.5 /  2   =1 .. XX .. .. .. .. .. ..
 4: Spike 1.2 Turin           1.0 /  2   10 .. .. XX .. .. .. .. ..
 5: Thinker                   1.0 /  2   01 .. .. .. XX .. .. .. ..
 6: Fruit 2.3.1               0.5 /  2   =0 .. .. .. .. XX .. .. ..
 7: bright-0.2c               0.0 /  2   00 .. .. .. .. .. XX .. ..
 8: Toga II 3.1.2SE JA Chekov 0.0 /  2   00 .. .. .. .. .. .. XX ..
 9: Naum 2.0                  0.0 /  2   00 .. .. .. .. .. .. .. XX
------------------------------------------------------------------
16 games: +7 =2 -7

Ratings of participants so far:

Code: Select all

Rank Name                        Elo    +    - games score oppo. draws 
   1 bright-0.3a                2789  145  135    16   63%  2706   13% 
   2 Rybka 1.0 Beta 32-bit      2780   93   87    38   67%  2668   24% 
   3 Thinker                    2773   90   89    42   51%  2760   12% 
   4 Toga II 3.1.2SE JA Chekov  2729   90   87    38   58%  2673   21% 
   5 Glaurung 2.0.1             2723   90   88    38   57%  2674   24% 
   6 Spike 1.2 Turin            2711   89   88    38   55%  2675   26% 
   7 bright-0.2c                2680   91   92    38   50%  2678   11% 
   8 Fruit 2.3.1                2632   88   91    38   42%  2683   21% 
   9 Naum 2.0                   2620   87   91    38   41%  2685   18% 

PGN at the end of the tourney.

After this, Bright 0.3a is the new best freeware! (on this list) Bright 0.3a is no doubt stronger than 0.2c but let's wait till the error bars aren't that high. Toga Chekov's material for activity approach seems to not work against Bright. However this approach works against Glaurung, but Glaurung defeats Bright on a direct match.

It would be interesting to make a three way match between Bright, Glaurung and Toga 3.1.2SE Chekov because I perceive some sort of Rock Paper Scissors effect.

Also, Rybka is over Thinker for now.

[hgm]

Well, considering the rating difference and the error bars, it would be more accurate to say that "Rybka has been slightly more lucky that Thinker, for now". You need about 100 times as many games to say anything about the sign of their rating difference with reasonable confidence (like twice as likely of being right than wrong)...

[Ovyron]

Thank you, I will remember to use the word "lucky" from now on

[Ovyron]

Bright 0.3a got a lot unlucky on the second and third rounds, and the rating dropped greatly, its drawn games percentage also went up so Bright 0.2c is still so far the king of decided games.

Cross-table:

Code: Select all

Bright Spark
2008.03.29 - 2008.04.03
                                 Score     brig Rybk Glau Thin Spik Toga brig Frui Naum
----------------------------------------------------------------------------------------
 1: bright-0.3a                15.5 / 32   XXXX =00= 00== 10=0 0110 1100 11=0 =1== 11==
 2: Rybka 1.0 Beta 32-bit       3.0 /  4   =11= XXXX .... .... .... .... .... .... ....
 3: Glaurung 2.0.1              3.0 /  4   11== .... XXXX .... .... .... .... .... ....
 4: Thinker                     2.5 /  4   01=1 .... .... XXXX .... .... .... .... ....
 5: Spike 1.2 Turin             2.0 /  4   1001 .... .... .... XXXX .... .... .... ....
 6: Toga II 3.1.2SE JA Chekov   2.0 /  4   0011 .... .... .... .... XXXX .... .... ....
 7: bright-0.2c                 1.5 /  4   00=1 .... .... .... .... .... XXXX .... ....
 8: Fruit 2.3.1                 1.5 /  4   =0== .... .... .... .... .... .... XXXX ....
 9: Naum 2.0                    1.0 /  4   00== .... .... .... .... .... .... .... XXXX
----------------------------------------------------------------------------------------
32 games: +12 =11 -9

Ratings of participants so far:

Code: Select all

   1 Rybka 1.0 Beta 32-bit      2776   90   84    40   68%  2664   25% 
   2 Thinker                    2773   88   86    44   52%  2752   14% 
   3 Toga II 3.1.2SE JA Chekov  2739   89   86    40   60%  2667   20% 
   4 Glaurung 2.0.1             2717   86   84    40   56%  2670   28% 
   5 Spike 1.2 Turin            2706   87   86    40   55%  2671   25% 
   6 bright-0.3a                2691   91   91    32   48%  2706   34% 
   7 bright-0.2c                2685   89   88    40   51%  2673   13% 
   8 Fruit 2.3.1                2631   85   87    40   43%  2678   25% 
   9 Naum 2.0                   2621   84   87    40   41%  2679   23% 

PGN with 32 games:
http://www.zshare.net/download/10005725646a298b/

Next engine to test: Junior 10. I've seen that it performs really poor under the Fischerclock, but I'll see how bad it does

[Dr.Wael Deeb]

Not because I am the author of the Bright's opening book,I have much better results with own opening book....I'll post later when the tournament is finished

[Ovyron]

After testing Junior I'll start testing these engines with own books. Bookless, manual book learning and 1 forced move have been fun but I'm considering giving these engines their chance to show their true potential