Stockfish Docs

Appearance

Setup ​

Software prerequisites ​

Python ​

If using easy_train.py then at least 3.7 is required.

Otherwise versions around 3.6 should also work, but updating is recommended.

Python 3.11 was also tested and works well.

A C++ compiler ​

If you're using easy_train.py then at least GCC 9.2 is required for compiling the data loader, Stockfish, and c-chess-cli. This a defensive version requirement as older versions were known to cause problems with Stockfish.

If you're not using easy_train.py then no automatic compilation will take place; use what you wish.

If you're on Windows the best way is to probably use MSYS2. It would also make it easy to install Make (next step).

Make ​

For compiling Stockfish and c-chess-cli. Not strictly necessary if you're not using easy_train.py, but recommended.

CMake ​

Necessary for compiling the data loader.

https://cmake.org/install/

Package dependencies ​

python -m venv trainer
pip install -r requirements.txt

PyTorch with CUDA 11.8 will be automatically installed, along with the matching CuPy version.

The data loader (if not using easy_train.py) ​

This requires a C++17 compiler and cmake.

Windows:

compile_data_loader.bat

Linux/Mac:

sh compile_data_loader.bat

Network training and management ​

Hard way: wiki

Easier way: wiki

Logging ​

TODO: Move to wiki. Add setup for easy_train.py

tensorboard --logdir=logs

Then, go to http://localhost:6006/

Automatically run matches to determine the best net generated by a (running) training ​

TODO: Move to wiki

python run_games.py --concurrency 16 --stockfish_exe ./stockfish.master --c_chess_exe ./c-chess-cli --ordo_exe ./ordo --book_file_name ./noob_3moves.epd run96

Automatically converts all .ckpt found under run96 to .nnue and runs games to find the best net. Games are played using c-chess-cli and nets are ranked using ordo. This script runs in a loop, and will monitor the directory for new checkpoints. Can be run in parallel with the training, if idle cores are available.

Thanks ​