Gauntlet Berserk 6

[Rebel]

Gauntlet Berserk 6 : estimated +114 elo

Code: Select all

Gambit Rating List
Running      : Gauntlet Berserk 6
Time Control : Time control : 40/120
Games        : 1000

Results from file gauntlet-berserk-6.pgn:

No. Name            Win Draw Loss Unf.  Score Games       %
-----------------------------------------------------------
  1 Berserk 6      +364 =477 -159   *0  602.5  1000   60.2%
  2 Igel 3.0.5      +38 =102  -60   *0   89.0   200   44.5%
  3 Ethereal 12.75  +33 =107  -60   *0   86.5   200   43.2%
  4 Pedone 3.1      +34 =100  -66   *0   84.0   200   42.0%
  5 Komodo 11       +29  =86  -85   *0   72.0   200   36.0%
  6 Seer 2.3.0      +25  =82  -93   *0   66.0   200   33.0%

Total Games:    1000
White Wins:      259 (25.9%)
Black Wins:      264 (26.4%)
Draws:           477 (47.7%)
Unfinished:        0 (0.0%)

Estimated elo gain for Berserk_6
Elo pool : 3351
Berserk 4.5.0 : 3298.0
Berserk_6 : 3412.3
Difference : 114.3

We try another 1000 games against an even stronger pool.

Elo pool : 3411

http://rebel13.nl/b/grl.htm

[matejst]

Great result from both Berserkers so far! I think that B6 is about the strength of Koivisto 6.16, with a bit better evaluation but a slower search.

[amanjpro]

Great result from both Berserkers so far! I think that B6 is about the strength of Koivisto 6.16, with a bit better evaluation but a slower search.

Berserk reports "shallower" depth than Koivisto, because it does a ton of extensions... To truly compare them you should probably want to compare seldepth or NPS

[matejst]

Dear Amanj,

I am not trying to be so precise. I noticed in my short test that K6.16 reports higher depths and that Berserk plays better equal endgames. That's all. But in general I am positively surprised by its game so far, although it was just a quick glance.

I read that you helped Jay with the NN, and I am glad there is so much cooperation among "young" -- so to speak -- developers.

[amanjpro]

Dear Amanj,

I am not trying to be so precise. I noticed in my short test that K6.16 reports higher depths and that Berserk plays better equal endgames. That's all. But in general I am positively surprised by its game so far, although it was just a quick glance.

Sorry, I wasn't trying to falsify the hypothesis, I was just pointing to the strength point of Berserk. On paper Zahak searches deeper than Berserk, but much narrower... And Berserk beats Zahak almost in every game

I read that you helped Jay with the NN, and I am glad there is so much cooperation among "young" -- so to speak -- developers.

It was nice from Jay to credit me, all I did was having developed Zahak, that he could use it to test his trainer
But sure, after all chess programming is a hobby, and cooperation is the key

[matejst]

It was nice from Jay to credit me, all I did was having developed Zahak, that he could use it to test his trainer
But sure, after all chess programming is a hobby, and cooperation is the key

Credits matter too, yes, it's kind, but then it is also appropriate. And then, I see a new community growing in the nicest way, on healthy bases. I now hope for an engine for end-users, not for TCEC. But I guess it will come soon.

[mehmet123]

Hello Ed. Why don' t you test Fire?
It's hard for me to understand why there is an air of protest against this chess engine among testers.

[Rebel]

Hello Ed. Why don' t you test Fire?
It's hard for me to understand why there is an air of protest against this chess engine among testers.

Basically (I think) you can divide computer chess lovers (regrading this topic) into 3 categories. What they have in common is the strongest playing strength possible, the differences are:

1. First group doesn't care about the origin of an engine even if it's an illegal clone or derivative breaking the GPL or another license.

2. Second group will not use illegal clone or derivatives but are using legal (those who keep the GPL) clones or derivatives for their testing, tournaments and rating lists.

3. A third group (mainly the programmers) don't like to play clones or derivatives because they don't want to compete against clones or derivatives at all. It's no fun at all to play a tournament or in a rating list against 3-4 Stockfish derivatives which will hurt their ranking, no fair competition. So far 70 chess programmers have agreed to a code of honor, made a statement about the origin of their engine with the promise to keep the stipulations as listed on - http://rebel13.nl/download/programmer-c ... honor.html In other words honesty, transparency and a healthy development / competition ambiance.

Both Houdini and Fire (by me) are considered as derivatives. They have no place in the GRL.

[Rebel]

Great result for Berserk 6, scored a rating of 3414.

Code: Select all

Gambit Rating List
Running      : gauntlet Berserk 6
Time Control : Time control : 40/120
Games        : 1000

Results from file gauntlet-berserk-6a.pgn:

No. Name               Win Draw Loss Unf.  Score Games       %
--------------------------------------------------------------
  1 Berserk 6         +240 =528 -232   *0  504.0  1000   50.4%
  2 Stockfish 9        +55 =105  -40   *0  107.5   200   53.8%
  3 RubiChess 2.2      +45 =112  -43   *0  101.0   200   50.5%
  4 Komodo 12          +55  =88  -57   *0   99.0   200   49.5%
  5 SlowChess 2.6      +39 =113  -48   *0   95.5   200   47.8%
  6 Koivisto 6.0-sse2  +38 =110  -52   *0   93.0   200   46.5%

Total Games:    1000
White Wins:      220 (22.0%)
Black Wins:      252 (25.2%)
Draws:           528 (52.8%)
Unfinished:        0 (0.0%)

Estimated ratings for this elo 3411 pool

   # PLAYER               :  RATING  POINTS  PLAYED   (%)
   1 Stockfish 9          :  3439.7   107.5     200    54
   2 RubiChess 2.2        :  3416.8   101.0     200    51
   3 Berserk 6            :  3413.3   504.0    1000    50
   4 Komodo 12            :  3409.8    99.0     200    50
   5 SlowChess 2.6        :  3397.6    95.5     200    48
   6 Koivisto 6.0-sse2    :  3388.8    93.0     200    47

[Rebel]

Current top-10

Code: Select all

Gambit Rating List         : Provisional

   # PLAYER                 :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D     L  D(%)
   1 Stockfish 21-10-06     :  3712.9   22.3  1593.0    2000    80      94  1212   762    26    38
   2 Komodo-Dragon 2.5      :  3648.2   17.7  1527.5    2200    69     100   998  1059   143    48
   3 Lc0 v28                :  3587.4   23.3   591.5    1000    59     100   335   513   152    51
   4 Ethereal 13.25-NNUE    :  3464.1    9.2  2049.0    4000    51     100  1101  1896  1003    47
   5 Koivisto 6.16          :  3452.3   10.2  1168.0    2600    45      92   536  1264   800    49
   6 SlowChess 2.7          :  3438.1   16.4  1177.0    2800    42      98   458  1438   904    51
   7 Berserk 6              :  3418.5   14.6  1106.5    2000    55      84   604  1005   391    50
   8 RubiChess 2.2          :  3412.6    8.4  1880.5    4200    45      98   876  2009  1315    48
   9 Igel 3.0.5             :  3363.4   12.5  1724.5    4500    38      57   682  2085  1733    46
  10 Pedone 3.1             :  3362.2   11.1  1942.5    4500    43      57   972  1941  1587    43