My strongest NNUE net (yet)

[ChickenLogic]

Quick test 30 + 0.2 of your net (stnnuecl1) against the previous best net known for me. Yours is a bit better:

Code: Select all

   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 stnnuecl1    : 2868.0   10.7    269.0     504   53.4%
   2 stnnue       : 2844.0   10.7    235.0     504   46.6%

Now testing it against the latest one of Additional Pylon halfkp 256.

Yeah, about that. I kinda have something to tell you.

You see, this Additional Pylon dude on discord (I hate him) is kinda the same person as me.

[Raphexon]

Quick test 30 + 0.2 of your net (stnnuecl1) against the previous best net known for me. Yours is a bit better:

Code: Select all

   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 stnnuecl1    : 2868.0   10.7    269.0     504   53.4%
   2 stnnue       : 2844.0   10.7    235.0     504   46.6%

Now testing it against the latest one of Additional Pylon halfkp 256.

Yeah, about that. I kinda have something to tell you.

You see, this Additional Pylon dude on discord (I hate him) is kinda the same person as me.

Trust no one, not even yourself.

[cdani]

Quick test 30 + 0.2 of your net (stnnuecl1) against the previous best net known for me. Yours is a bit better:

Code: Select all

   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 stnnuecl1    : 2868.0   10.7    269.0     504   53.4%
   2 stnnue       : 2844.0   10.7    235.0     504   46.6%

Now testing it against the latest one of Additional Pylon halfkp 256.

Yeah, about that. I kinda have something to tell you.

You see, this Additional Pylon dude on discord (I hate him) is kinda the same person as me.

!!!

[cdani]

I have done a longer time control test. 600 seconds (10 minutes) + 2 seconds against latest Stockfish dev. So maybe 80 elo less than Stockfish.

Code: Select all

 # PLAYER         : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2896.3    9.9    203.5     333   61.1%
   2 stnnuecl1    : 2815.7    9.9    129.5     333   38.9%

Games:
https://1drv.ms/u/s!Aj8wlSmr_Hq3if4nhbX ... g?e=FDQtrt

[lkaufman]

I have done a longer time control test. 600 seconds (10 minutes) + 2 seconds against latest Stockfish dev. So maybe 80 elo less than Stockfish.

Code: Select all

 # PLAYER         : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2896.3    9.9    203.5     333   61.1%
   2 stnnuecl1    : 2815.7    9.9    129.5     333   38.9%

What value are you using for slowmover? My best result was with 70, with 60 close behind.

[cdani]

What value are you using for slowmover? My best result was with 70, with 60 close behind.

Default one.

[cdani]

I have done a longer time control test. 600 seconds (10 minutes) + 2 seconds against latest Stockfish dev. So maybe 80 elo less than Stockfish.

Code: Select all

 # PLAYER         : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2896.3    9.9    203.5     333   61.1%
   2 stnnuecl1    : 2815.7    9.9    129.5     333   38.9%

What value are you using for slowmover? My best result was with 70, with 60 close behind.

I'm repeating the test with Slow Mover 60. Tomorrow the results.

[Jesse Gersenson]

Does this run in linux?

[cdani]

I have done a longer time control test. 600 seconds (10 minutes) + 2 seconds against latest Stockfish dev. So maybe 80 elo less than Stockfish.

Code: Select all

 # PLAYER         : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2896.3    9.9    203.5     333   61.1%
   2 stnnuecl1    : 2815.7    9.9    129.5     333   38.9%

What value are you using for slowmover? My best result was with 70, with 60 close behind.

I'm repeating the test with Slow Mover 60. Tomorrow the results.

Here the results with Slow Mover 60:

Code: Select all

   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2886.2    9.8    220.0     377   58.4%
   2 stnnuecl1    : 2825.8    9.8    157.0     377   41.6%

So as told is better, maybe 60 elo from Stockfish dev.

The games:
https://1drv.ms/u/s!Aj8wlSmr_Hq3if4rM1y ... w?e=0baeAp

[cdani]

I have done also a test in multithreading, 3 threads as the computers have 6 cores.
Somehow the short time control is very bad, but at long time control is ok:

Code: Select all

Stc 50 + 0.3 3 threads: 
   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2916.6   11.2    290.0     437   66.4%
   2 stnnuecl1    : 2795.4   11.2    147.0     437   33.6%

Ltc 200+0.5 3 threads
   # PLAYER       : RATING  ERROR   POINTS  PLAYED    (%)
   1 st200617     : 2887.8    9.0    207.5     352   58.9%
   2 stnnuecl1    : 2824.2    9.0    144.5     352   41.1%

Now I will test the newest nets of ChickenLogic vs the one of Gekkehenker.