A TensorFlow-based project to make a deep learning model learn to play the game Pong based on the paper "Playing Atari with Deep Reinforcement Learning" by Deepmind (Mnih et al., 2019).
The game environment is implemented with pygame and reconstructs the behavior of the Atari game.
- Deep Deterministic Policy-Gradient (DDPG) Methods
- Double Deep Q Networks (DDQN)
- Deep Q Networks (DQN)
- Experience Replay and Epsilon Greedy
Preferred Requirements: Python 3.9
The package can be installed with hatch or a virtual environment in the project's root directory:
python -m venv .venv
source .venv/bin/activate
pip install -e .Run one of the experiments, e.g. DDQN:
python experiments/train_ddqn.pyNow you can locate the trained model and load it within main.py to see the performance versus a simple AI algorithm or play against the AI agent.