huskyth / CPPZero

AlphaZero Gomoku

A multi-threaded implementation of AlphaZero

Features

• Easy Free-style Gomoku
• Multi-threading Tree/Root Parallelization with Virtual Loss and LibTorch
• Gomoku, MCTS and Network Infer are written in C++
• SWIG for Python C++ extension
• Update 2019.7.10: Supporting Ubuntu and Windows
• Update 2022.4.4: Re-compile with CUDA 11.6/PyTorch 1.10/LibTorch 1.10(Pre-cxx11 ABI)/SWIG 4.0.2

Args

Edit config.py

Packages

• Python 3.6+
• PyGame 1.9+
• CUDA 10+
• PyTorch 1.1+
• LibTorch 1.1+ (Pre-cxx11 ABI)
• SWIG 3.0.12+
• CMake 3.8+
• MSVC14.0+ / GCC6.0+

Run

# Compile Python extension
mkdir build
cd build
cmake .. -DCMAKE_PREFIX_PATH=path/to/libtorch -DPYTHON_EXECUTABLE=path/to/python -DCMAKE_BUILD_TYPE=Release
make -j10

# 我的mac下使用这个编译通过了，添加了DCMAKE_OSX_ARCHITECTURES
cmake .. -DCMAKE_PREFIX_PATH=/home/husky/Downloads/libtorch-cxx11-abi-shared-with-deps-2.2.1+cu121/libtorch -DPYTHON_EXECUTABLE=/home/husky/miniconda3/envs/mcts/bin/python -DCMAKE_BUILD_TYPE=Release

# https://blog.csdn.net/Felaim/article/details/105832560
# 可以查看PYTHON_LIBRARIES和PYTHON_INCLUDE_DIRS
# kaggle中这样可以编译通过

# PYTHON_INCLUDE_DIR
# >>> from distutils.sysconfig import get_python_inc
# >>> print(get_python_inc())
# /home/husky/miniconda3/envs/mcts/include/python3.11

# PYTHON_LIBRARY
# >>> import distutils.sysconfig as sysconfig
# >>> print(sysconfig.get_config_var('LIBDIR'))
# /home/husky/miniconda3/envs/mcts/lib

!cmake .. -DCMAKE_PREFIX_PATH=/opt/conda/lib/python3.10/site-packages/torch/share/cmake -DPYTHON_EXECUTABLE=/opt/conda/bin/python  -DCMAKE_BUILD_TYPE=Release -DPYTHON_LIBRARY=/opt/conda/lib -DPYTHON_INCLUDE_DIR=/opt/conda/include/python3.10

# export TORCH_CUDA_ARCH_LIST="8.0 8.6 8.9 9.0"
# 崩潰日志
# 1. txt位置修改

# swig -c++ -python example.i
# python setup.py build_ext --inplace

# Run
cd ../test
python learner_test.py train # train model
python learner_test.py play  # play with human

Pre-trained models

Trained 2 days on GTX TITAN X (similar to GTX1070)

See GitHub Release: https://github.com/hijkzzz/alpha-zero-gomoku/releases

GUI

References

1. Mastering the Game of Go without Human Knowledge
2. Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm
3. Parallel Monte-Carlo Tree Search
4. An Analysis of Virtual Loss in Parallel MCTS
5. A Lock-free Multithreaded Monte-Carlo Tree Search Algorithm
6. github.com/suragnair/alpha-zero-general