Durak Reinforcement Learning
Training RL agents via self-play to master the imperfect-information card game Durak using a C++ game engine and Python training pipeline.
C++ Python Reinforcement Learning
What We’re Building
| Type | Learning project |
| Focus | Reinforcement learning and agentic modelling |
| Final Product | Playable game against trained RL agents |
| Repo | durak-reinforcement-learning ↗ |
Architecture
- C++ Game Engine — Durak rules, game state, legal actions, step function. Exposed to Python via pybind11.
- Python Training — PyTorch neural networks trained through self-play using the C++ engine for fast simulation.
- Frontend — UI to play against trained agents (TBD).
Skills / Want to Learn
Reinforcement Learning Self-Play C++ pybind11 PyTorch Game Theory Imperfect Information Games