Zahak, a GoLang based chess engine

[Wolfgang]

Running soon for CEGT-40/4:
https://cegt.forumieren.com/t1601-testi ... 8-0nn#3090

[Rebel]

Part II

Estimated elo gain for Zahak 8.0 : +193

Code: Select all

Gambit Rating List
Running      : Gauntlet Zahak 8.0
Time Control : Time control : 40/120
Games        : 1000

Results from file gauntlet-zahak2.pgn:

No. Name                 Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------------
  1 Zahak 8.0           +321 =306 -373   *0  474.0  1000   47.4%
  2 Weiss 1.4            +80  =69  -51   *0  114.5   200   57.2%
  3 Stash 31.0           +78  =68  -54   *0  112.0   200   56.0%
  4 Demolito 2021-09-30  +81  =52  -67   *0  107.0   200   53.5%
  5 Beef 0.3.6           +71  =56  -73   *0   99.0   200   49.5%
  6 Counter 4.0          +63  =61  -76   *0   93.5   200   46.8%

Total Games:    1000
White Wins:      318 (31.8%)
Black Wins:      376 (37.6%)
Draws:           306 (30.6%)
Unfinished:        0 (0.0%)

Estimated elo gain for Zahak_8.0
Elo pool : 3142
Zahak 7.0 : 2933.0
Zahak_8.0 : 3126.7
Difference : 193.7

[amanjpro]

Part II

Estimated elo gain for Zahak 8.0 : +193

Code: Select all

Gambit Rating List
Running      : Gauntlet Zahak 8.0
Time Control : Time control : 40/120
Games        : 1000

Results from file gauntlet-zahak2.pgn:

No. Name                 Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------------
  1 Zahak 8.0           +321 =306 -373   *0  474.0  1000   47.4%
  2 Weiss 1.4            +80  =69  -51   *0  114.5   200   57.2%
  3 Stash 31.0           +78  =68  -54   *0  112.0   200   56.0%
  4 Demolito 2021-09-30  +81  =52  -67   *0  107.0   200   53.5%
  5 Beef 0.3.6           +71  =56  -73   *0   99.0   200   49.5%
  6 Counter 4.0          +63  =61  -76   *0   93.5   200   46.8%

Total Games:    1000
White Wins:      318 (31.8%)
Black Wins:      376 (37.6%)
Draws:           306 (30.6%)
Unfinished:        0 (0.0%)

Estimated elo gain for Zahak_8.0
Elo pool : 3142
Zahak 7.0 : 2933.0
Zahak_8.0 : 3126.7
Difference : 193.7

Thanks for testing again... Im happy either way... I was expecting it to be close to Combusken and that is what your test shows

[AdminX]

Can you provide a link to the NNUE file that you are using for testing Zahak 8.0? I am not sure I am using the correct file.

Thanks

[Graham Banks]

Can you provide a link to the NNUE file that you are using for testing Zahak 8.0? I am not sure I am using the correct file.

Thanks

Pretty sure that it's embedded in the engine exe.

[AdminX]

Can you provide a link to the NNUE file that you are using for testing Zahak 8.0? I am not sure I am using the correct file.

Thanks

Pretty sure that it's embedded in the engine exe.

Hmm, In the engine dialog there is a option to link it to a NNUE file (Banksia GUI). So I downloaded the Default NNUE (default.nn) from Zahak's site on Github. I tested with and without the 'default.nn' being set using 1 thread and Fritz 17 won 4 games in a row which is not what I was expecting.

[AdminX]

Maybe it was just luck last night, so I am trying it again.

[Wolfgang]

Pretty sure that it's embedded in the engine exe.

+1

[amanjpro]

Maybe it was just luck last night, so I am trying it again.

The default.nn happened to be exactly what is included (if you were looking into the master branch)

Safest is to use no network, and leave it with the included network. The option is there so, if I trained a better net people can start using it without the need to wait for a recompile/new release. Or if one wishes, one can load weaker nets for human play (that is what Scally is doing)

Thanks for testing Zahak

[AdminX]

Maybe it was just luck last night, so I am trying it again.

The default.nn happened to be exactly what is included (if you were looking into the master branch)

Safest is to use no network, and leave it with the included network. The option is there so, if I trained a better net people can start using it without the need to wait for a recompile/new release. Or if one wishes, one can load weaker nets for human play (that is what Scally is doing)

Thanks for testing Zahak

Thanks!