Hello, I am trying to implement maia-chess to my game:

[Descartador]

I am developing an educative game about chess and my idea was to present characters that play more human-like. I am using Ren'py on MAC but I am unsure how to go about implementing maia-chess to the game, any suggestions on how to go about this in a smart way?

Thank you!

[reidmcy]

In the code for the paper we used python chess with lc0, see here: https://github.com/CSSLab/maia-chess/blob/master/move_prediction/maia_chess_backend/model_loader.py. There is a slightly better version in the newer repo: https://github.com/CSSLab/maia-chess/blob/master/move_prediction/maia_chess_backend/model_loader.py, but if you want it to not rely on subprocesses you'll want to use something based on the original TF code from either our releases our the Leela training code https://github.com/LeelaChessZero/lczero-training.

[lwtw123456]

Hi, I came across this issue and thought my project might be useful as a reference.

I recently integrated bqrio/maia3-onnx into Chess Titans, the chess game bundled with Windows 7 x64. I replaced the game's built-in AI with Maia3 for move selection, translating Chess Titans' internal board/move representation to the inputs and policy outputs expected by the ONNX model.

My implementation is obviously specific to Windows/Chess Titans, so it isn't directly reusable for Ren'Py on macOS, but the Maia3 ONNX integration, board encoding, and move-policy mapping parts may still be useful as a reference.

Repository: https://github.com/lwtw123456/ChessTitans-Maia3

Hope it helps!