Gull 3 Linux+Syzygy and Fathom released.

[Dan Cooper]

anton.mihailov@gmail.com

At least get a dialogue going so he knows to contact you when it's time for stability testing.

[velmarin]

Longer days to note this, but never with other code...(Original is different)
Gull delivers its Tower, at any position with these three parts...
It is very curious.


Analysis by Gull 3 sy.05 x64:

1.Te7
# (#0) Profundidad: 1/1 00:00:00

[d]3k4/8/8/8/8/3N4/2BKR3/8 w - - 0 0

[basil00]

At least get a dialogue going so he knows to contact you when it's time for stability testing.

Thanks, email sent.

The basic TODO list before stage 2:

• Figure out best compiler to build Gull (MinGW (gcc) or MSVC?), build options, etc.?
• Testing, testing, testing.
• Demonstrate an elo gain over non-syzygy Gull 3 on Windows

[Dan Cooper]

At least get a dialogue going so he knows to contact you when it's time for stability testing.

Thanks, email sent.

The basic TODO list before stage 2:

• Figure out best compiler to build Gull (MinGW (gcc) or MSVC?), build options, etc.?
• Testing, testing, testing.
• Demonstrate an elo gain over non-syzygy Gull 3 on Windows

I'm doing some testing now. Will be done in about a week.

TODO list:
King safety improvements
bench command

[basil00]

King safety improvements
bench command

Improvements (+elo) are most welcome. A bench command is also something we need.

[stuwph]

Longer days to note this, but never with other code...(Original is different)
Gull delivers its Tower, at any position with these three parts...
It is very curious.


Analysis by Gull 3 sy.05 x64:

1.Te7
# (#0) Profundidad: 1/1 00:00:00

[d]3k4/8/8/8/8/3N4/2BKR3/8 w - - 0 0

in fact probing dtz doesn't work at all
spent some hours to find a reason but being unsuccessful

[basil00]

in fact probing dtz doesn't work at all
spent some hours to find a reason but being unsuccessful

Gull currently just plays the TB move, whatever it is. In the case of a win, it is whatever move has the least DTZ value that preserves the win. In this case the move appears to be Re7, which seemingly blunders the rook and looks very unnatural. However, if black takes the rook Gull will win the KNBvK endgame, so the WLD50 value is perfectly preserved.

[stuwph]

yes, I know this seemingly immanet syzygy behaviour ... but this is not the case
at least with a mingw 5.3 compile under Windows

the position is mate in #7
e.g. 1.Nc5 / 1.Nc4 / 1. Nf4 / Ba4

no other engine with syzygy 5- or 6-men plays 1.Re7


with this Gull selfplay goes

1.Re7 #(#0) Kc8 # (#0)
2.Rc7+ #(#0) Kb8 # (#0)
3. Kc1 #(#0) Ka8 # (#0)
and so on ...
evaluation stays always at #0 and not thinking (Fritz15 gui)

in Arena 3.51 or cli it's different but confused too

switching the hash to other sizes like 64, 128, 1024 leads to other evaluations or even exceptions of the engine

think there is a hidden bug, maybe only in windows

[Dan Cooper]

Ran 2 gauntlets with Gull and Gullzygy.
40/1 tc
+11 elo

Code: Select all

   # PLAYER                   : RATING  ERROR    POINTS  PLAYED     (%)   CFS(next)
   1 Stockfish 031516 BMI2    : 2520.7   22.2     436.0     600   72.7%      97    
   2 Komodo 9.42 C10          : 2491.3   20.8     415.5     600   69.3%     100    
   3 Houdini 4                : 2388.9   20.1     335.0     600   55.8%     100    
   4 Gull 051116 BMI2         : 2352.9    8.5    1918.0    3300   58.1%      96    
   5 Gull 3 x64 BMI2          : 2341.8    8.7    1872.5    3300   56.7%      63    
   6 Fire 4                   : 2337.9   19.9     292.0     600   48.7%     100    
   7 Equinox 3.30             : 2290.7   20.0     252.5     600   42.1%     100    
   8 Fizbo 1.7                : 2241.4   20.1     213.0     600   35.5%      92    
   9 Andscacs 0.86            : 2219.7   20.0     196.5     600   32.8%      97    
  10 Texel 1.06a45            : 2190.7   21.9     175.5     600   29.3%      68    
  11 Nirvana 2.2 POPCNT       : 2182.8   22.0     170.0     600   28.3%      72    
  12 Sting SF 6.7             : 2173.2   21.7     163.5     600   27.3%      62    
  13 Hannibal 1.5             : 2168.0   22.2     160.0     600   26.7%     ---    

White advantage = 39.28 +/- 3.56
Draw rate (equal opponents) = 46.95 % +/- 0.70

 4) Gull 051116 BMI2      2352.9 :   3300 (+1245,=1346,-709),  58.1 %

    vs.                          :  games (    +,    =,   -),   (%) :    Diff,    SD, CFS (%)
    Stockfish 031516 BMI2        :    300 (   19,  128, 153),  27.7 :  -167.8,  12.2,    0.0
    Komodo 9.42 C10              :    300 (   35,  121, 144),  31.8 :  -138.4,  11.8,    0.0
    Houdini 4                    :    300 (   59,  146,  95),  44.0 :   -36.0,  11.2,    0.1
    Fire 4                       :    300 (   61,  197,  42),  53.2 :   +15.1,  11.2,   91.2
    Equinox 3.30                 :    300 (  104,  150,  46),  59.7 :   +62.3,  11.2,  100.0
    Fizbo 1.7                    :    300 (  141,  101,  58),  63.8 :  +111.5,  11.1,  100.0
    Andscacs 0.86                :    300 (  156,  106,  38),  69.7 :  +133.2,  11.4,  100.0
    Texel 1.06a45                :    300 (  166,   98,  36),  71.7 :  +162.2,  12.2,  100.0
    Nirvana 2.2 POPCNT           :    300 (  154,  107,  39),  69.2 :  +170.2,  12.2,  100.0
    Sting SF 6.7                 :    300 (  178,   92,  30),  74.7 :  +179.7,  11.9,  100.0
    Hannibal 1.5                 :    300 (  172,  100,  28),  74.0 :  +185.0,  12.3,  100.0

 5) Gull 3 x64 BMI2       2341.8 :   3300 (+1196,=1353,-751),  56.7 %

    vs.                          :  games (    +,    =,   -),   (%) :    Diff,    SD, CFS (%)
    Stockfish 031516 BMI2        :    300 (   15,  132, 153),  27.0 :  -179.0,  12.4,    0.0
    Komodo 9.42 C10              :    300 (   31,  116, 153),  29.7 :  -149.6,  11.4,    0.0
    Houdini 4                    :    300 (   59,  148,  93),  44.3 :   -47.2,  11.2,    0.0
    Fire 4                       :    300 (   55,  187,  58),  49.5 :    +3.9,  11.3,   63.5
    Equinox 3.30                 :    300 (   88,  161,  51),  56.2 :   +51.1,  11.1,  100.0
    Fizbo 1.7                    :    300 (  142,  107,  51),  65.2 :  +100.4,  11.4,  100.0
    Andscacs 0.86                :    300 (  139,  111,  50),  64.8 :  +122.1,  11.3,  100.0
    Texel 1.06a45                :    300 (  160,   99,  41),  69.8 :  +151.1,  12.2,  100.0
    Nirvana 2.2 POPCNT           :    300 (  176,   93,  31),  74.2 :  +159.0,  12.1,  100.0
    Sting SF 6.7                 :    300 (  160,  105,  35),  70.8 :  +168.5,  12.0,  100.0
    Hannibal 1.5                 :    300 (  171,   94,  35),  72.7 :  +173.8,  12.6,  100.0

[carldaman]

in fact probing dtz doesn't work at all
spent some hours to find a reason but being unsuccessful

Gull currently just plays the TB move, whatever it is. In the case of a win, it is whatever move has the least DTZ value that preserves the win. In this case the move appears to be Re7, which seemingly blunders the rook and looks very unnatural. However, if black takes the rook Gull will win the KNBvK endgame, so the WLD50 value is perfectly preserved.

Why not run a search before, instead of auto-playing the TB move? Other top engines do that and achieve more realistic EG play.