Usage

This guide covers all the ways to use the UNO Card Game RL system.

Game Modes

Human vs AI

The primary game mode where you play against a trained AI agent.

python uno_gui.py

Gameplay:

Launch the GUI
Select your preferred AI model from the dropdown
Click “Play vs AI”
Wait for cards to be dealt
Click on highlighted (playable) cards to play them
If no playable card, click the deck to draw
First to empty their hand wins!

AI vs AI (Spectator Mode)

Watch two AI agents battle each other:

python uno_gui.py
# Click "AI vs AI"

This mode is useful for:

Observing AI strategies
Debugging agent behavior
Entertainment

Model Battle Arena

The battle arena allows you to:

Compare multiple models head-to-head
Run batch evaluations (10-1000 games)
Track statistics (wins, win rates, average turns)
Export results to CSV

python model_battle_gui.py

Features:

2-4 player support
Select different models for each player
Real-time statistics
Progress bar for batch runs

Multiplayer Mode

Play with 3-4 AI agents simultaneously:

python multiplayer_gui.py --players 4

Command-Line Interface

Quick Game

For a fast text-based game:

python run.py

Model Comparison

Compare models programmatically:

python compare_models.py --games 100

Configuration

Environment Variables

Set these environment variables to customize behavior:

# Disable GPU (CPU only)
export CUDA_VISIBLE_DEVICES=""

# Set random seed
export UNO_SEED=42

Config File

Edit config.py to modify:

Model paths
Training hyperparameters
Game settings

# config.py
sb3_models = {
    "selfplay_champion": {
        "name": "Self-Play Champion",
        "path": "models/selfplay_champion.zip",
        "win_rate": 0.70
    },
    # ... other models
}

Keyboard Shortcuts

GUI Controls

Key	Action	Context
ESC	Return to menu	During game
Space	Draw card (if no playable)	Your turn
R	Restart game	Game over
Q	Quit	Any screen

Mouse Controls

Left Click: Select cards, buttons
Hover: View card details
Scroll: (Battle arena) Scroll through history

Output Files

The system generates several output files:

Comparison Results

Located in comparison_results/:

trained_models_YYYYMMDD_HHMMSS.csv - Batch evaluation results
existing_models_YYYYMMDD_HHMMSS.csv - Pre-trained model comparison

Training Logs

Located in logs/:

evaluations.npz - Evaluation metrics during training
*/events.out.tfevents.* - TensorBoard logs

View with TensorBoard:

tensorboard --logdir logs/

Assets

Located in assets/:

*.csv - Historical training data
*.png - Training plots (if generated)