How keithlee96/pluribus-poker-ai works
Project type and purpose
• This is a full-stack Python toolkit for building and using a poker AI based on Counterfactual Regret Minimisation (MCCFR).
• It provides a core library (poker_ai), a CLI (poker_ai cluster|train|play), batch apps (dataset/visualisation), a live web visualiser, and research scripts.
File structure
• Top level: Dockerfiles (Dockerfile, ParentDockerfile), CI/CD configs (.travis.yml, GitHub Actions), docs, requirements.txt, setup.py.
• poker_ai/: the core Python package, subdivided into:
• ai/ – MCCFR runner, agent abstractions, multiprocess manager & workers
• games/ – state & player models (e.g. ShortDeckPokerState)
• poker/ – engine, table, pot, deck, hand evaluation
• terminal/ – CLI runner for text UI
• utils/ – I/O, logging, random seed
• applications/: standalone scripts and servers
• strategy_dag_visualisation/ – dataset generator
• visualisation/ – a Flask/WebSocket backend + Vue.js frontend
• research/: exploratory scripts
• test/: unit & functional tests
README for additional insights
• CLI commands: cluster, train start|resume, play, each invoking core library modules.
• Docker-based CI builds the environment, runs pytest.
• The visualiser runs a local Flask server exposing WebSockets; frontend is a webpack/Vue single-page app.
• Data artifacts: card_info_lut.joblib, CSV dataset, strategy files, results.yaml.