Deep Reinforcement Learning Algorithms

Testing Deep RL Algorithms with Lunar Lander and Atari Games.

Pong	Breakout

To Install:

1. git clone repo
2. Create virtual environment
3. pip install -r requirements.txt

   (if using testStable.py file, install requirementsStable.txt)

To run:

Run main.py

Options:

All options for testing are changed within settings.json file

To change algorithm change 'rlOption' from 1-5

• 1 is DQN (set 'ddqn' variable to either true or false for Double DQN).
• 2 is Expected SARSA
• 3 is REINFORCE
• 4 is Vanilla Actor Critic
• 5 is A2C (note: this one does not perform correctly)

To change game, change 'Game' variable:

• 1 is Pong
• 2 is Breakout
• 3 is Space Invaders

To change Model, use either 'Advanced' or 'Basic':

• 'Advanced' Model is Convolutional Layer Pytorch network.
• 'Basic' Model, when this is selected the game will automatically be the Lunar Lander Environment.

Results:

• Run plot.py to then visualize test run.
• playModel.py can also be used to play trained model weights, this must be selected however in the playModel file.