nnue-pytorch VS Ceres

There's extremely detailed information on how Stockfish's neural network evaluation works, but none of this will tell you the why of why the engines' assessment changed: https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md

If you want a readable explanation of all the details, this document is phenomenal.

Getting SOTA in chess would be earth-shattering, especially since Stockfish has now adopted very light-weight NNs (called NNUE) and has doubled down on alpha-beta search, regaining the upper hand against A0 style programs.

Stockfish NNUE is deep network. You can find out more about it's architecture and internal working here:

https://github.com/glinscott/nnue-pytorch/blob/master/docs/n...

It's pretty interesting read.

[0] https://tests.stockfishchess.org/tests/view/60dae5363beab81350aca077 [1] https://nextchessmove.com/dev-builds [2] https://stockfishchess.org/blog/2021/stockfish-13/ [3] https://lczero.org/blog/2021/06/the-importance-of-open-data/ [4] https://github.com/official-stockfish/Stockfish/commit/e8d64af1 [5] https://github.com/glinscott/nnue-pytorch/ [6] https://stockfishchess.org/get-involved/

https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md There is some info there.

something like this exists already it's called Ceres https://github.com/dje-dev/Ceres/releases/tag/v0.94

That being said it is not clear which approach is better e.g. a very smart but relatively slow evaluation function (Leela) or SFs approach to use a relatively dumb eval function with a more sophisticated search. It is pretty clear that Leelas search can be improved (Check https://github.com/dje-dev/Ceres for example)