/239as-rl-proj

Primary LanguagePython

UCLA ECE 239AS Reinforcement Learning Final Project

This repository contains the source code for the implementations we used to evaluate three deep Q-learning algorithms:

  1. Deep Q-learning (DQN)
  2. Double Deep Q-learning (Double DQN or DDQN)
  3. Dueling DQN

Additionally, the repository hosts the LaTeX source for the various reports we had to produce, and the poster we created to present a summary of our work.

For the algorithm implementations, we leveraged example code from the Tensorpack repository, with minimal modifications to fit our goals and limited compute resources.

Team members

  • Ryan Chau
  • My-Quan Hong
  • Nathan Kang
  • Chris Munoz

Getting started

To clone the repository, run the following command:

git clone https://github.com/cmunozcortes/239as-rl-proj.git project

Requirements/Dependencies

Instructions for installing dependencies and required packages can be found here.

Training

To train a model to play breakout, run the following commands:

cd src
python DQN.py --env breakout.bin --algo <algorithm>

where <algorithm> can be DQN, Double, Dueling.

Evaluation

To evaluate a trained model for 100 episodes, run the following command:

python DQN.py --env breakout.bin --task eval --load <trained_model> --algo <algorithm>