Helllo Vivien,
the lerning code is the the learner from Stockfish (including Stockfish Qsearch). I only created a generator for the training data, which helped me to create several 100 millions of Nemorino evals and store it in the sgen-format expected by the learner.
Nemorino's net is worse than the original Nets from Stockfish. Running Nemorino with a Stockfish net will gain further Elo (in selfplay something like 30-50 Elo).
I will try to create my own learner in the future (as this is an interesting challenge) but this needs time. My main motivation is trying to understand and learn how everything works, and I learned a lot the last weeks. If I will ever have an own learner I surely will give you access to it.
Best regards,
Christian
Florentino wrote: ↑Mon Sep 28, 2020 10:11 pm
I just released version 6 of Nemorino!
Main new feature is the support for NNUE networks, which improves Nemorino's playing strength a lot!
Nemorino comes with it's own network. The network format is slightly different than the original format from Stockfish as I removed the 64 "BONA_PIECE_ZERO" input parameters from the weights file and from the inference code (which is an adapted copy of the code from Stockfish). Nevertheless Nemorino can use the network files from Stockfish as well - it will simply create a converted copy on first use.
Additionally as it's more than 2 years since the last release, this version contains a lot of minor improvements and bugfixes.
Nemorino was on my list of engines for which to produce a net. Now I don’t have to. It’s good to see you back in the engine game.
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".
Florentino wrote: ↑Sun Oct 04, 2020 4:15 pm
No - as soon as SyzygyPath is set, only the external Syzygy-Files are used. Do you think such a mixed scenario makes sense?
Florentino wrote: ↑Mon Sep 28, 2020 10:11 pm
I just released version 6 of Nemorino!
Main new feature is the support for NNUE networks, which improves Nemorino's playing strength a lot!
Nemorino comes with it's own network. The network format is slightly different than the original format from Stockfish as I removed the 64 "BONA_PIECE_ZERO" input parameters from the weights file and from the inference code (which is an adapted copy of the code from Stockfish). Nevertheless Nemorino can use the network files from Stockfish as well - it will simply create a converted copy on first use.
Additionally as it's more than 2 years since the last release, this version contains a lot of minor improvements and bugfixes.