Deep Reinforcement Learning on Atari Games

This repo is a continuous effort to show the results of different reinforcement learning algorithms on Atari games. This work follows the works from the respective papers as well as the book by Maxim Lapan.

Install

This project requires python and the following python libraries installed.

1. torch
2. tensorboardX
3. gym
4. numpy
5. cv2

Run

1. Clone or download the repo using https://github.com/intelaashish/Comparing_Reinforcement_Learning_algorithms_on_Atari_Games.git
2. Navigate to the directory for the algorithm that you want to train on.
3. Run the training file. E.g.,

      python training_pong.py
   
   

Algorithms (to be) Covered:

• Deep Q Network (DQN)
• N-steps DQN
• Double DQN
• Dueling DQN
• Reinforce
• A2C
• A3C
• PPO
• TRPO
• DDPG
• TD3