README.md

Simple DQN implementation for Atari's discrete action space.
Algorithm from: https://www.nature.com/articles/nature14236.

Simply run python train.py.

• The agent should learn to play Breakout in about 2 hrs.
• Expert level performance takes a few more hours.
• Also provided trained model checkpoints for breakout and pong after about a million gradient steps

Code by Vishwas Sathish
Code references:

• https://pytorch.org/tutorials/intermediate/reinforcement_q_learning.html
• https://github.com/KaleabTessera/DQN-Atari/tree/master
• https://github.com/diegoalejogm/deep-q-learning/tree/master
• https://github.com/jzhanson/breakout-demo/tree/master