Deep Reinforcement Learning

DQN, Double DQN, Dueling DQN, Prioritized Experience Replay, Atari, keras, tensorflow, gym

Various Deep Reinforcement Learning techniques applied to different environments. Will be updated with new algorithms and environments often!

Just added (03/15/2019) is a general version of DQN with Double DQN and Dueling DQN compatibility. This version is located in the gym directory.

Requirements

• python 3.6
• keras
• tensorflow
• gym
• gym['atari']
• numpy
• OpenCV
• random
• collections

Getting Started

Running this code is very simple. Make sure that you have the above requirements taken care of, then download the two python files. In the command line, or any python editor change directory to either gym or Atari folder where the python files are located and type:

python dqn.py --train_dqn

or for DDQN:

python ddqn.py --train_ddqn

If you can also render the atari environment by typing:

python ddqn.py --train_ddqn --render

To specify a certain number of episodes (example: 10) to run (50,000 default) use:

python ddqn.py --train_ddqn --episodes 10

Finally to test the final model after training, you can pass the test_dqn (or test_ddqn) flag:

python ddqn.py --test_ddqn --render --episodes 10

This code makes use of an atari wrapper code that can be found here

Priortized experience replay version of this code will be uploaded shortly!

Recent Additions:

• (03/15/2019): DQN for gym environments with Double and Dueling DQN

If you have any questions or comments, don't hesitate to send me an email! I am looking for ways to make this code even more computationally efficient. Contact info can be found here

---