NNUE Research Project

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
jshriver
Posts: 1342
Joined: Wed Mar 08, 2006 9:41 pm
Location: Morgantown, WV, USA

Re: NNUE Research Project

Post by jshriver »

Ferdy wrote: Tue Apr 06, 2021 5:11 am Thanks for sharing. Do you have info what options were used to generate these bins? What stockfish version?
I've been keep a record and will add the gensfens options sometime soon. Switched to a new configuration that is supposed to be base on the what the current best nnue nets have used as input. Should start uploading my first batch this weekend.

The files currently available used this:

gensfen set_recommended_uci_options book noob_3moves.epd random_move_maxply 4 write_minply 14 random_move_count 13 random_multi_pv_diff 200 depth ${DEPTH} loop ${LOOP} save_every 100 output_file_name ${FNAME}

The ones I'm building now are using this:

gensfen depth ${DEPTH} loop ${LOOP} random_multi_pv 4 random_multi_pv_diff 100 set_recommended_uci_options sfen_format binpack book noob_3moves.epd seed ${RANDOM}${RANDOM} output_file_name ${FNAME}

They are generated using nodechips current stockfish master.
Ferdy
Posts: 4833
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: NNUE Research Project

Post by Ferdy »

jshriver wrote: Thu Apr 08, 2021 1:21 am
Ferdy wrote: Tue Apr 06, 2021 5:11 am Thanks for sharing. Do you have info what options were used to generate these bins? What stockfish version?
I've been keep a record and will add the gensfens options sometime soon. Switched to a new configuration that is supposed to be base on the what the current best nnue nets have used as input. Should start uploading my first batch this weekend.

The files currently available used this:

gensfen set_recommended_uci_options book noob_3moves.epd random_move_maxply 4 write_minply 14 random_move_count 13 random_multi_pv_diff 200 depth ${DEPTH} loop ${LOOP} save_every 100 output_file_name ${FNAME}

The ones I'm building now are using this:

gensfen depth ${DEPTH} loop ${LOOP} random_multi_pv 4 random_multi_pv_diff 100 set_recommended_uci_options sfen_format binpack book noob_3moves.epd seed ${RANDOM}${RANDOM} output_file_name ${FNAME}

They are generated using nodechips current stockfish master.
Thanks, this is convenient to genererate nets at 500M. random_move_maxply 4 (default=24) implies that there are only few random moves that might be done and random_move_count 13 (default=5) implies that more random moves will be saved in each game, this would mean that more positions will be saved in the middle and ending phases.
Madeleine Birchfield
Posts: 512
Joined: Tue Sep 29, 2020 4:29 pm
Location: Dublin, Ireland
Full name: Madeleine Birchfield

Re: NNUE Research Project

Post by Madeleine Birchfield »

Rebel wrote: Sat Mar 13, 2021 11:39 am I am looking for more engines that moved to NNUE.

Anyone?

The data so far is too less for conclusions.
Would be interesting to return to this project, now that time has passed and a lot more engines (new ones especially) have started to use NNUE. Koivisto, Berserk, Zahek, Arasen, Scorpio, and so on