Hyperion Chess Engine

Hyperion is a MCTS-NN based chess engine built primarily in C++. The Engine is still in development with plans to add reinforcement learning soon!

Project Structure (Two Builds)

There are two different and separate builds for this project:

1. Python (Training)
   • The Python code (using PyTorch) is strictly used for training to create the neural network models.
   • To set up the Python environment, install the requirements by running:

     pip install -e .

   • Navigate to the training directory and run:

     python train.py

2. C++ (Engine and Inference)
   • The C++ code is the actual chess engine. It uses the neural network files trained by the Python component (saved in data/completed_models) for its inferences during searches.

Perft test results:

• Nodes per second: 4.39454e+07 (43.9 Million NPS)

Prerequisites: LibTorch (For C++ Engine)

Because this engine uses a Neural Network, it requires LibTorch (the C++ API for PyTorch) to build.

1. Download LibTorch from the PyTorch website.
   • Linux: Make sure to select the cxx11 ABI version.
   • Windows: Download the Release version. IMPORTANT: Do not use MinGW on Windows. You should instead use MSVC.
2. Extract the downloaded folder somewhere on your computer. You will need the path to this folder for the build steps below.

How to build C++ Engine?

CRITICAL: It is super important that you build the project in Release mode, not Debug! The Debug mode does not work with LibTorch and will cause errors.

1. Open a terminal in the hyperion directory.

2. Next, create and cd into the build directory:

   mkdir build
   cd build

3. Configure the project with CMake, making sure to provide the path to your extracted LibTorch folder and specifying the Release build type:

   cmake -DCMAKE_PREFIX_PATH="C:/absolute/path/to/your/libtorch" -DCMAKE_BUILD_TYPE=Release ..

4. Still in the build directory, compile the engine (Release config is passed to cmake build for MSVC):

   cmake --build . --config Release

5. Run the executable:

   .\bin\Release\HyperionEngine.exe

   (Note: The exact path might be .\bin\HyperionEngine.exe on non-MSVC builds, such as Linux.)

How to build C++ Engine with performance enhancements:

1. Open a terminal in the hyperion directory.

2. Next, cd into the build directory (hyperion/build/).

3. Run CMake with optimization flags, the LibTorch path, and Release mode:

   cmake -DCMAKE_PREFIX_PATH="C:/absolute/path/to/your/libtorch" -DHYPERION_ENABLE_BMI2=ON -DCMAKE_BUILD_TYPE=Release ..

4. Still in the build directory, build using multiple cores for a faster compile time:

   cmake --build . --config Release -j 8

5. Run the executable:

   .\bin\Release\HyperionEngine.exe

Tasks Completed

• Chess Logic Library
• Basic Monte-Carlo Tree Search Implementation
• Initial Neural Network Creation (Supervised Learning)
• Initial Neural Network Implementation
• Neural Network Self-Play (Reinforcement Learning)
• Refinements/Advanced Features