pong-dqn-rl

Deep Dueling Neural Network Reinforcement Learning for Pong.

This is an implementation of the Duel Double DQN algorithm in PyTorch for solving the OpenAI GYM Atari Pong environment which learns to play at an impressive level within approximately 1,000 episodes.

Requirements

• Python 3.10
• Python tools;
  • Poetry for environment management
  • Pyenv for managing Python versions
• Python packages;
  • torch for neural networks
  • gym for RL environments
  • opencv-python for image processing
  • configparser for configuration
  • black, isort & mypy for formatting

Method

Run the commands in the repository Makefile for training, formatting and testing the code.

Experiments

All experiments are configured in the config.ini file, ran in pong_dqn_rl/training.py with models, parameters and logs being saved to the models directory.

Resources

Playing Atari with Deep Reinforcement Learning

Dueling Network Architectures for Deep Reinforcement Learning