Welcome to the Simple Tag Competition! This project uses the Simple Tag environment from PettingZoo's MPE.

In this competition, you will train the predator agent only. Your predator will be evaluated against a public reference prey provided in this repository.

• Simple Tag: A cooperative/competitive multi-agent environment where three predators try to catch a prey
• Three predators chase a single prey in a continuous 2D space
• Predators are rewarded for catching prey; the prey is rewarded for avoiding capture

You must submit via Pull Request from your fork:

1. One Python file: agent.py - Contains your agent implementation
2. Model weights (optional): Any .pth files for your neural networks

Your submission should follow this structure:

submissions/<your_username>/
├── agent.py           # Required: Your predator implementation
└── predator_model.pth # Optional: Your predator neural network weights

Your agent.py must implement the following class (predator only):

class StudentAgent:
    def __init__(self):
        """
        Initialize your predator agent.
        """
        pass
    
    def get_action(self, observation, agent_id: str):
        """
        Get action for the given observation.
        
        Args:
            observation: Agent's observation from the environment
            agent_id: Unique identifier for this agent instance
            
        Returns:
            action: Action to take in the environment
        """
        pass

See template/agent.py for a complete template.

1. Fork this repository
2. Create your submission folder: submissions/<your_username>/
3. Add your agent.py (and optional .pth files)
4. Create a Pull Request to the main repository
5. Wait for automatic evaluation - results will appear on the leaderboard

• Your predator is evaluated against a public reference prey
• Each PR triggers automatic evaluation via GitHub Actions
• Results are published to the leaderboard immediately
• Note: PRs are not merged - they are only used for evaluation

• You may use any RL algorithm (DQN, PPO, SAC, etc.)
• You may train your agents however you like
• You may use pre-trained models
• You can submit multiple times (new PRs will update your score)
• Do not modify files outside your submission folder
• Submit one agent.py file and optionally any .pth model files

# Clone the repository
git clone https://github.com/nathanael-fijalkow/simple-tag-competition.git
cd simple-tag-competition

# Install dependencies
pip install -r requirements.txt

python test_agent.py submissions/<your_username>/agent.py

Check the live leaderboard at: https://nathanael-fijalkow.github.io/simple-tag-competition/

The leaderboard shows:

• Student username
• Predator score (average reward vs public prey)
• Submission timestamp
• Ranking