2048 Game + DQN Reinforcement Learning Agent

This project contains:

• A clean 2048 game environment (like OpenAI Gym style)
• A Deep Q-Network (DQN) agent built using PyTorch
• A training loop to learn how to play 2048
• Modular, easy-to-extend code

---

🚀 How to Run

1. Install Requirements

pip install -r requirements.txt

2. Train the DQN Agent

python dqn_2048_game.py

Training will print average score every 100 episodes.

---

📦 Project Structure

.
├── dqn_2048_game.py     # Full Game + DQN training
├── requirements.txt      # Dependencies
├── README.md             # You're reading it!

---

🧠 How It Works

• Environment: 2048 board (4x4), states are normalized grids
• Actions: 0 = Up, 1 = Down, 2 = Left, 3 = Right
• Rewards: Based on increase in board tile sum; small penalty for invalid move
• Algorithm: Standard DQN with:
  • Experience Replay
  • Target = Bellman Equation (reward + discounted future reward)

---

📈 Future Upgrades

• Add Target Networks (DQN stabilization)
• Use DDQN (Double DQN)
• Prioritized Experience Replay
• Add TensorBoard logging
• Play with a trained agent (inference mode)

---

📜 Requirements

See requirements.txt below.

---

requirements.txt

torch
numpy

(You only need torch and numpy for this basic version!)

---

🏆 Credits

Built for fun and learning reinforcement learning principles with a classic game: 2048.

---

✨ Enjoy training your AI to beat 2048!