New release: rofChade 3.0

[Ronald]

I’ve been doing a lot of NN training for TCEC season 22, which eventually resulted in an unbelievable 4th place in the premier division! The infra final for 3th place against Lc0 showed that there is still a mayor gap between the top 3 engines and the rest including rofChade, but still an amazing result.

After the tournament I didn’t have much time to work on rofChade anymore, so it took some time to release the first NN version of rofChade, but here it is!

The current NN network architecture is HALFKA (2×256)x32x32x1. For now, this still gives the best results for rofChade. The network is trained with 2.8B by rofChade generated positions. Around 600M of them are FRC positions. The positions have been rescored a few times with intermediate networks to simulate Reinforcement learning.

The NN code is implemented from scratch, the matrix calculations are heavily influenced by the Cfish implementation. For the training of the networks an older version of the nnue-pytorch trainer from Stockfish is used as base.

To be able to generate FRC positions for training, rofChade 3.0 also supports FRC and DFRC chess variants.

I also upgraded the version of the fathom library, so that 7-men syzygy tablebases are also supported (although I haven’t been able to test if it works).

A special thanks to this version goes to the following persons/groups:

• Ronald de Man for his matrix calculations in Cfish and the syzygy tablebase library
  The group that develops the nnue-pytorch trainer (and their interesting discussions on discord)
  Andrew Grant for sharing his experience with his FRC implementation and perft test data. It helped me to discover some unexpected FRC bugs
  Jon Dart for the fathom library and updates on it
  Frank Quisinsky for letting intermediate rofChade versions play in it’s tournament
  IpmanChess for testing some intermediate rofChade versions
  Graham banks and the rest of the CCRL team for doing a great testing job for so long
  The TCEC and CCC team to organize great tournaments
  Everybody else who enjoys (computer) chess!

[bastiball]

Congrats for the released!

[Frank Quisinsky]

Code: Select all

    # Player                      :      Elo  Games  Score%   won  draw  lost  Points  Draw%  Error   OppAvg   OppE   OppD

  08. rofChade 2.321 NN dev       :  3368.25    750    65.6   285   414    51   492.0   55.2  16.12  3241.50  11.59   43.5
  --. rofChade 2.317 NN dev       :  3345.14   1200    65.5   451   670    79   786.0   55.8  13.06  3220.07   9.80   40.0
  --. rofChade 2.313 NN dev       :  3291.69   1230    60.8   410   675   145   747.5   54.9  12.42  3206.71   9.95   41.0
  --. rofChade 2.310 NN dev       :  3255.73   1200    59.3   370   683   147   711.5   56.9  12.35  3183.20  10.74   40.0
  --. rofChade 2.3                :  3169.06   1200    53.8   302   687   211   645.5   57.3  11.48  3139.82  10.24   40.0
 

Hi Ronald,

you are welcome!
I have a lot of fun with your engine.
Thank you that I could participate on your very hard work around the dev versions in the last months.

Move average in comparsion to other TOP-10 engines are really good.
Playing style for the dev. versions are the same (never the program lost his own face).
Your engine plays mid-game, endgame very constantly.
At any time slightly aggressive!

The community will have a lot of fun with rofChade!!

Als ik weer in Nederland ben, gaan we frikandeln eten (uitjes, frietjes en majo, ketchup ... alles moet erin + Heineken). Had een Nederlandse friendin toen hij jong was. Misschien spreek ik better Nederlands dan Engels.

Andrew Grant is a great helper for many programmers. I read that and hear that again and again. My disagreement with him is realy bad for me but often I am ... "to crazy". If he read that ... hello Andrew ... have a nice weekend!

Go rofChade go ...

Best
Frank

Code: Select all

109. rofChade 2.3                   (302+,   687=,   211-)  53.8%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         94     1200
           white won         89      204
           black won         89       98
          white draw         98      342
          black draw        100      345
          white loss         92       54
          black loss         82      157
            all wins         89      302     1200      25.2%
           all draws         99      687     1200      57.2%
           all loses         84      211     1200      17.6%

110. rofChade 2.310 NN dev          (370+,   683=,   147-)  59.3%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         92     1200
           white won         82      243
           black won         85      127
          white draw         98      330
          black draw         97      353
          white loss        100       27
          black loss         87      120
            all wins         83      370     1200      30.8%
           all draws         97      683     1200      56.9%
           all loses         89      147     1200      12.2%

111. rofChade 2.313 NN dev          (410+,   675=,   145-)  60.8%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         92     1230
           white won         83      262
           black won         89      148
          white draw        102      327
          black draw         93      348
          white loss         93       26
          black loss         88      119
            all wins         85      410     1230      33.3%
           all draws         97      675     1230      54.9%
           all loses         89      145     1230      11.8%

112. rofChade 2.317 NN dev          (451+,   670=,   79-)  65.5%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         92     1200
           white won         82      293
           black won         90      158
          white draw        103      285
          black draw         93      385
          white loss         95       22
          black loss         85       57
            all wins         85      451     1200      37.6%
           all draws         97      670     1200      55.8%
           all loses         88       79     1200       6.6%

113. rofChade 2.321 NN dev          (285+,   414=,   51-)  65.6%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         93      750
           white won         84      192
           black won         93       93
          white draw        102      180
          black draw         95      234
          white loss         96        3
          black loss         89       48
            all wins         87      285      750      38.0%
           all draws         98      414      750      55.2%
           all loses         89       51      750       6.8%

[Gabor Szots]

Ronald, may I ask you about the differences between v2.321 and v3.0? To make my point clear, do you think that their results may be safely combined or shall we treat them separately?

[Frank Quisinsky]

Hi Ronald,

BTW:
Alex is working on a "new" Mephisto module.
Have a look in Wasp news on my website (here you can find pictures).

I am reading on your site that you have interest to work on a Rasp version.
Important here is that people can later adjust the Elo strength.

Wasp will be inside the project.
rofChade would be great!

If you have interest on it, let me know ... I will send you the mail address from Alex!

Best
Frank

[Ronald]

Ronald, may I ask you about the differences between v2.321 and v3.0? To make my point clear, do you think that their results may be safely combined or shall we treat them separately?

Version 3.0 contains a different net, but it was only marginally stronger in my tests (2/3 elo..). The other thing changed is the support for 7-men tablebases.

[Ronald]

As noted by Guenther Simon the SSSE3 and SSE2 versions were compiled with hardware popcount on. I replaced the old zips on the website with new ones that include the correct compiles. The name of the downloaded zip starts with "pop".

If you already downloaded rofChade and want to use 1 of the 2 versions, you should download rofChade again.