Pong DQN

An implementation of a Deep Q-Network (DQN) agent that learns to play Atari Pong using reinforcement learning.

Based on: OpenAIPong-DQN

Files

• Pong_DQN.ipynb: Jupyter Notebook containing the code for training, evaluating the agent, and visualizing the CNN outputs.
• requirements.txt: List of required Python packages.
• saved_models/:
  • pong-cnn-860.pkl: Trained model weights after 860 episodes.
• assets/video_recordings/:
  • rl-video-episode-846.mp4: Video of the agent achieving a high score.

Usage

1. Install Dependencies

   pip install -r requirements.txt

2. Run the Notebook

   Open Pong_DQN.ipynb in Jupyter Notebook or JupyterLab and run the cells to train the agent, load the pre-trained model, or visualize the CNN outputs.

Results

• The agent demonstrates high proficiency in playing Pong after training.
• A gameplay video is available in the assets/video_recordings/ folder.
• The notebook includes an implementation for visualizing the CNN outputs, providing insights into what the agent has learned.