regi_py

This repo contains a C++ implementation of the game mechanics of the card game regicide, with wrappers for Python and WASM, and reinforcement learning models trained via PyTorch.

Try the WASM version of the game here: once loaded, the game (including neural-net bots) should run fully in the browser, without requiring any connection to a server.

Installation

Install this package via Python's pip tool on the command line:

# activate anaconda/uv/whatever for Python env
git clone https://github.com/ahgamut/regi-py
cd regi-py
python3 -m pip install -e .

Building and testing

The package ships a C++ extension (regi_py.core) that must be compiled for your interpreter. Build it in-place and run the test suite with:

# from a fresh virtualenv
python -m pip install --upgrade pip setuptools wheel
python -m pip install "pybind11>=3"
python -m pip install -e .[dev] --no-build-isolation   # builds the extension + test deps
pytest -q

Re-run pip install -e . --no-build-isolation after any change to the C++ sources under src/regi_py/core/ to rebuild the extension before testing.

python -m pytest webdriver/tests   # NOT bare `pytest webdriver/tests`

Viewing a basic simulation

Run the driver.py to see a basic command-line simulation of the game, where each player randomly selects a valid move to play.

# install the package first
python driver.py

To use custom strategies available in the package, specify them with --add-bot:

python driver.py --help # view options for bots
python driver.py --add-bot damage --add-bot preserve

Playing with a bot

To play locally, run webdriver/driver.py to play a 2-player game on your local server:

# install the package first
cd webdriver
python driver.py \
    -n 1 \ # one human player needs to connect
    --add-bot damage # add the 'damage' bot
# go to http://localhost:8888 in your browser

You can also build the WASM version and run it locally. To use the neural nets locally, you'll need to train them by yourself.

Adding your own strategies

Subclass the BaseStrategy class with your own implementations that select what attack/defense moves to make.