Just stumbled upon an open-source free engine

[Laskos]

Googling for some open-source engines, I came across some communists, they were bad-mouthing capitalists for reasons beyond me. These communists had bad manners and were speaking through a broken Babelfish. Then they offered me an engine to put to the test.

I took the 5 times World Champion, state-of-the-art undisputed leader Rybka 4 to make a quick job out of these che-guevaras. Here is the result (using LittleBlitzer 2.5):

Code: Select all

Games Completed = 10000 of 60000 (Avg game length = 5.975 sec)
Settings = RR/16MB/500ms+50ms/M 1000000cp for 1000 moves, D 200 moves/EPD:D:\LittleBlitzer\Hyatt4000.epd
Time = 71937 sec elapsed, 532331 sec remaining
 1.  Ivanhoe B49jA_1-cpu_x64  	5277.0/10000	3820-3266-2914  	(L: m=3266 t=0 i=0 a=0)	(D: r=1758 i=637 f=395 s=104 a=20)	(tpm=49.9 d=10.5 nps=1019769)
 2.  Rybka 4_x64              	4723.0/10000	3266-3820-2914  	(L: m=3820 t=0 i=0 a=0)	(D: r=1758 i=637 f=395 s=104 a=20)	(tpm=52.2 d=7.5 nps=62088)

Then, hoping to trash them away using Elostat:

Code: Select all

    Program                            Score        %    Av.Op.  Elo    +   -    Draws

  1 Ivanhoe B49jA_1-cpu_x64        : 5277.0/10000  52.8   3190   3210    6   6   29.1 %
  2 Rybka 4_x64                    : 4723.0/10000  47.2   3210   3190    6   6   29.1 %

again

20 +/-6 Elo points 95% confidence for these communists against the undisputed leader Rybka 4.

My last hope was in the LOS, Rybka 4 may have less points, but a better LOS.

LOS = 99.9999999977% for communists

Please, can someone help me to have Rybka 4 at a LOS>99% against these devils?

Kai

[zullil]

Code: Select all

Games Completed = 10000 of 60000 (Avg game length = 5.975 sec)
Settings = RR/16MB/500ms+50ms/M 1000000cp for 1000 moves, D 200 moves/EPD:D:\LittleBlitzer\Hyatt4000.epd
Time = 71937 sec elapsed, 532331 sec remaining
 1.  Ivanhoe B49jA_1-cpu_x64  	5277.0/10000	3820-3266-2914  	(L: m=3266 t=0 i=0 a=0)	(D: r=1758 i=637 f=395 s=104 a=20)	(tpm=49.9 d=10.5 nps=1019769)
 2.  Rybka 4_x64              	4723.0/10000	3266-3820-2914  	(L: m=3820 t=0 i=0 a=0)	(D: r=1758 i=637 f=395 s=104 a=20)	(tpm=52.2 d=7.5 nps=62088)

Kai

Does Rybka still provide incorrect depth and/or nps data? Or am I simply not understanding the output you posted?

It seems that Ivanhoe averaged depth 10.5 and 1000000 nps while Rybka averaged depth 7.5 and only 62000 nps. Something seems very wrong here.

Can anyone clarify meaning of the data Kai posted? Thanks.

[Laskos]

Code: Select all

Games Completed = 10000 of 60000 (Avg game length = 5.975 sec)
Settings = RR/16MB/500ms+50ms/M 1000000cp for 1000 moves, D 200 moves/EPD:D:\LittleBlitzer\Hyatt4000.epd
Time = 71937 sec elapsed, 532331 sec remaining
 1.  Ivanhoe B49jA_1-cpu_x64  	5277.0/10000	3820-3266-2914  	(L: m=3266 t=0 i=0 a=0)	(D: r=1758 i=637 f=395 s=104 a=20)	(tpm=49.9 d=10.5 nps=1019769)
 2.  Rybka 4_x64              	4723.0/10000	3266-3820-2914  	(L: m=3820 t=0 i=0 a=0)	(D: r=1758 i=637 f=395 s=104 a=20)	(tpm=52.2 d=7.5 nps=62088)

Kai

Does Rybka still provide incorrect depth and/or nps data? Or am I simply not understanding the output you posted?

It seems that Ivanhoe averaged depth 10.5 and 1000000 nps while Rybka averaged depth 7.5 and only 62000 nps. Something seems very wrong here.

Can anyone clarify meaning of the data Kai posted? Thanks.

Yes, being state-of-the-art, in computer chess is better to have "a fully knowledgeable engine" like Rybka, so the nodes are x(15-18) and depth is +3. Hope my test is very wrong...

Kai

[zullil]

so the nodes are x(15-18) and depth is +3. Hope my test is very wrong...

Kai

I did not realize that Rybka still gives incorrect search data. Thanks for the clarification.

[alpha123]

so the nodes are x(15-18) and depth is +3. Hope my test is very wrong...

Kai

I did not realize that Rybka still gives incorrect search data. Thanks for the clarification.

It counts nodes only for white and divides that by 7, so multiplying it by 14 will get you roughly the real node count. So Rybka really averaged about 868000 NPS.
It subtracts 2 or 3 from the depth (usually 3), so Rybka really got an average depth of about 10.5.

Peter