Komodo 2.01 is out!

[Laskos]

Gauntlets finished:

Code: Select all

    Program                            Score     %    Av.Op.  Elo    +   -    Draws

  Komodo64 2.01 64 bit           : 1213.5/3100  39.1   3200   3123   10  10   28.7 %

  1 Houdini 1.5a x64               : 463.0/620  74.7   3123   3309   27  26   22.3 %
  2 Deep Rybka 4.1 x64             : 397.5/620  64.1   3123   3222   23  23   32.1 %
  3 Ivanhoe B47cBx64-1             : 382.0/620  61.6   3123   3204   23  23   30.6 %
  4 Stockfish 2.1 JA 64bit         : 342.5/620  55.2   3123   3158   23  23   28.2 %
  5 Critter 1.01 64-bit            : 301.5/620  48.6   3123   3112   23  23   30.2 %

and

Code: Select all

    Program                            Score     %    Av.Op.  Elo    +   -    Draws

  Komodo64 1.3 JA                 : 821.0/3000  27.4   3200   3030   11  11   26.6 %

  1 Houdini 1.5a x64               : 487.0/600  81.2   3030   3282   30  30   18.3 %
  2 Deep Rybka 4.1 x64             : 453.0/600  75.5   3030   3224   26  26   27.3 %
  3 Ivanhoe B47cBx64-1             : 453.0/600  75.5   3030   3224   26  26   27.0 %
  4 Stockfish 2.1 JA 64bit         : 399.0/600  66.5   3030   3147   24  24   32.0 %
  5 Critter 1.01 64-bit            : 387.0/600  64.5   3030   3132   24  24   28.3 %

+93 +/- 15 Elo points (95% confidence) improvement. Somewhat smaller than in self-play, which is to be expected. Curious if my results will apply to "real" games, as you said, the programs scale differently to larger TC.

Kai

[Sarciness]

Nearly 100 elo. Damn impressive!!

[Dr.Wael Deeb]

Nearly 100 elo. Damn impressive!!

....to say the least

[stevenaaus]

I've added per-game time control to Scid vs. PC's tournament feature if anyone wants to mess around. (Komodo's per-move time control still seems broke)

At the moment it's alpha code, for UCI only... I think i'm doing it right. Code is in svn.

[Adam Hair]

Here is a partial result from CCRL 40/4 1 CPU testing for Komodo 2.01 and Stockfish 2.1.1:

Code: Select all

 5 Houdini 1.5a 64-bit                      3268   15   14  1973   73%  3090   28%
17 Rybka 4.1 64-bit                         3208   21   21   910   75%  3027   33%
20 Stockfish 2.1.1 64-bit                   3190   39   38   242   70%  3051   40%
22 Stockfish 2.0.1 64-bit                   3188   20   20  1128   77%  2976   28%
24 Rybka 3 64-bit                           3179    9    8  6001   74%  2994   30%
25 Critter 1.01 64-bit                      3173   20   20  1056   77%  2961   28%
32 Komodo 2.01 64-bit                       3157   39   38   231   65%  3054   36%
35 Critter 0.90 64-bit                      3147   19   18  1209   75%  2959   30%
60 Komodo 1.3 64-bit                        3087   12   12  2422   55%  3046   39%
66 Naum 4.2 64-bit                          3084   13   13  2022   57%  3039   41%

[Don]

Komodo seems to be crashing a bit. (I haven't tried the new version)

How does Komodo work with the Shredder Classic GUI?

Yes, Komodo appears to be crashing on occasion and I'm trying to fix that. The problem is that I cannot make it crash on Linux and I don't develop or test in windows. I do have a windows dual boot laptop but it's not crashing in arena. Or if it is I don't see where it is being reported.

It's hard to debug something that is not producing errors in my own tests. So I have asked a few people to help me debug and I'm working on the problem.

Don

regardless of any small issues/problems,
a big congrats is due to you (and Larry) for your tremendous effort on this excellent engine!

With the help of Dann Corbit I think the problem has been solved but we are now testing to verify. We found a bug in the routine that tests whether a move that was not generated by the normal move generator is legal or not and in some cases a pawn would move to the 8th rank and not promote. This breaks an implied contract in the program and causes others to be broken as a result.

Don

[Laskos]

Here is a partial result from CCRL 40/4 1 CPU testing for Komodo 2.01 and Stockfish 2.1.1:

Code: Select all

 5 Houdini 1.5a 64-bit                      3268   15   14  1973   73%  3090   28%
17 Rybka 4.1 64-bit                         3208   21   21   910   75%  3027   33%
20 Stockfish 2.1.1 64-bit                   3190   39   38   242   70%  3051   40%
22 Stockfish 2.0.1 64-bit                   3188   20   20  1128   77%  2976   28%
24 Rybka 3 64-bit                           3179    9    8  6001   74%  2994   30%
25 Critter 1.01 64-bit                      3173   20   20  1056   77%  2961   28%
32 Komodo 2.01 64-bit                       3157   39   38   231   65%  3054   36%
35 Critter 0.90 64-bit                      3147   19   18  1209   75%  2959   30%
60 Komodo 1.3 64-bit                        3087   12   12  2422   55%  3046   39%
66 Naum 4.2 64-bit                          3084   13   13  2022   57%  3039   41%

Thanks, CCRL 40/4 gives 70 +/- 38 Elo points (95%) improvement, CEGT 40/20 gives 110 +/- 27 points (95%), maybe they will converge to something inside mine 93 +/- 15 (95%) result, I am really curious if my result is valid.

Kai

[Don]

Here is a partial result from CCRL 40/4 1 CPU testing for Komodo 2.01 and Stockfish 2.1.1:

Code: Select all

 5 Houdini 1.5a 64-bit                      3268   15   14  1973   73%  3090   28%
17 Rybka 4.1 64-bit                         3208   21   21   910   75%  3027   33%
20 Stockfish 2.1.1 64-bit                   3190   39   38   242   70%  3051   40%
22 Stockfish 2.0.1 64-bit                   3188   20   20  1128   77%  2976   28%
24 Rybka 3 64-bit                           3179    9    8  6001   74%  2994   30%
25 Critter 1.01 64-bit                      3173   20   20  1056   77%  2961   28%
32 Komodo 2.01 64-bit                       3157   39   38   231   65%  3054   36%
35 Critter 0.90 64-bit                      3147   19   18  1209   75%  2959   30%
60 Komodo 1.3 64-bit                        3087   12   12  2422   55%  3046   39%
66 Naum 4.2 64-bit                          3084   13   13  2022   57%  3039   41%

Thanks, CCRL 40/4 gives 70 +/- 38 Elo points (95%) improvement, CEGT 40/20 gives 110 +/- 27 points (95%), maybe they will converge to something inside mine 93 +/- 15 (95%) result, I am really curious if my result is valid.

Kai

We have thousands of games that says it's 100 ELO, but that is at time controls that are around 40 seconds per game on very fast hardware (overclocked i7 6 core.)

YMMV

Don

[Aser Huerga]

CCRL 40/40 1CPU partial results:

Code: Select all

1   Komodo 2.01 64-bit     -3161- 17.5/38
2   Rybka 4.1 64-bit        3212  6.5/8
3   Houdini 1.5a 64-bit     3264  6.0/8
4   Stockfish 2.1.1 64-bit  3188  4.5/8
5   Critter 1.01 64-bit     3176  2.5/7
6   Naum 4.2 64-bit         3098  1.0/7

Code: Select all

1   Komodo 2.01 64-bit     -3177- 22.0/31
2   Spike 1.4 Leiden        3020  3.0/6
3   Gull 1.1 64-bit         3036  2.0/7
4   Protector 1.4.0 64-bit  3026  1.5/6  33.00
5   Hiarcs 13.2             3012  1.5/6  33.00
6   Spark 1.0 64-bit        3018  1.0/6

4 or 5 crashes till now (XP 64 bits)

[Don]

CCRL 40/40 1CPU partial results:

Code: Select all

1   Komodo 2.01 64-bit     -3161- 17.5/38
2   Rybka 4.1 64-bit        3212  6.5/8
3   Houdini 1.5a 64-bit     3264  6.0/8
4   Stockfish 2.1.1 64-bit  3188  4.5/8
5   Critter 1.01 64-bit     3176  2.5/7
6   Naum 4.2 64-bit         3098  1.0/7

Code: Select all

1   Komodo 2.01 64-bit     -3177- 22.0/31
2   Spike 1.4 Leiden        3020  3.0/6
3   Gull 1.1 64-bit         3036  2.0/7
4   Protector 1.4.0 64-bit  3026  1.5/6  33.00
5   Hiarcs 13.2             3012  1.5/6  33.00
6   Spark 1.0 64-bit        3018  1.0/6

4 or 5 crashes till now (XP 64 bits)

Hopefully the crash problem is solved - as soon as we prove that we will be releasing an update.

Don