jAIpur is a reinforcement learning (RL) environment and research toolkit for the board game Jaipur. It features a custom OpenAI Gymnasium-compatible environment, Stable Baselines3-style agent training (with MIMO support via @adysonmaia/sb3-plus), and is organized for future agent evaluation and human play.

• Jaipur Gym Environment:
  Implements all rules, tokens, and actions for Jaipur with multi-discrete action spaces and rich observations.
• Agent Training:
  Train agents using Stable Baselines3-style APIs and sb3-plus for multi-output (MIMO) policy support.
  Training is highly configurable via YAML config files.
• Extensible:
  Designed for easy addition of:
  • Agent-vs-agent or self-play evaluation scripts (coming soon)
  • Human-vs-agent CLI or GUI play (coming soon)

1. Clone this repository:

   git clone https://github.com/nicholasg1997/jAIpur.git
   cd jAIpur

2. Install dependencies:

   pip install -r requirements.txt

   • Requires Gymnasium, Stable Baselines3, and sb3-plus (by @adysonmaia), which provides MIMO policy support.

• All core environment and game rules are defined in the configs/ folder.
• The default RL hyperparameters are set in configs/ppo_configs.yaml.
• You can specify a different config directory at runtime to override defaults.

To train an agent with the default configuration:

python scripts/train_agent.py

• The training script loads the default config from configs/ppo_configs.yaml.
• To use a custom config, pass the directory or file location as an argument (see script for details).

Trained models are saved to the models/ directory by default.

• Test Agent vs. Itself:
  Scripts to evaluate a trained agent against itself or other agents are coming soon.
• Play Against the Agent:
  Play Jaipur against your trained agent via CLI or GUI (in development).

jaipur_rl/
  envs/         # Gymnasium environment
  game/         # Jaipur game logic
  training/     # Training utilities, callbacks, schedulers, trainer
configs/        # Game and RL configs (ppo_configs.yaml, game_configs.py, etc)
scripts/        # Training entry points
models/         # Saved agents

• Jaipur board game by Space Cowboys
• Stable Baselines3
• Gymnasium
• sb3-plus by @adysonmaia for multi-output policy support

Questions, suggestions, or want to contribute?
Open an issue or pull request on GitHub!