Quickstart
This guide will help you get started with UNO Card Game RL in just a few minutes.
Playing Your First Game
Play Against AI
Launch the main GUI and play against trained AI agents:
python uno_gui.py
Controls:
Select a model from the dropdown (e.g., “Best Recurrent PPO”)
Click “Play vs AI” to start
Click on playable cards (highlighted) to play them
Watch the AI take its turns
Watch AI vs AI
Observe trained models compete against each other:
python uno_gui.py
# Then click "AI vs AI"
Model Battle Arena
Compare multiple models in batch battles:
python model_battle_gui.py
Features:
Select 2-4 players
Choose different models for each player
Run batch evaluations (10-1000 games)
Export results to CSV
Available Models
Our trained models with their approximate win rates:
Model |
Win Rate |
Description |
|---|---|---|
Self-Play Champion ⭐ |
70%+ |
Best model, self-play trained |
Best Recurrent PPO |
60% |
LSTM-based, highest tested |
Optimal Recurrent PPO |
59% |
Hyperparameter optimized |
SB3 Recurrent PPO |
57% |
Standard SB3 training |
Best PPO |
53% |
MLP policy |
DQN |
48% |
Value-based |
Random Agent |
25% |
Baseline |
Running Training
Train Your Own Agent
Quick training with PPO:
python train_rl.py --algorithm ppo --timesteps 100000
Train with Self-Play (Recommended)
For the best results, use self-play training:
python training/train_selfplay.py --mode selfplay --timesteps 1000000
Evaluate a Model
Compare your trained model against baselines:
python compare_models.py --model models/your_model.zip --games 100
Command Reference
Command |
Description |
|---|---|
|
Main game GUI |
|
Model comparison arena |
|
3-4 player multiplayer |
|
Quick text-based game |
|
Batch model evaluation |
|
Train new model |
|
Self-play training |
What’s Next?
GUI Guide - Detailed GUI documentation
Training - Training your own agents
Algorithms - Understanding the RL algorithms
API Reference - API reference for developers