This is the corresponding GitHub page for a case study done on hyperparameter optimization for deep reinforcement learning using Atari Breakout as a study environment. We aimed to figure out the performance benefits of hyperparameter optimization in the case of a Deep Q-Learning (DQN) based reinforcement learning algorithm. We provide code for model training and hyperparameter tuning in this GitHub repository.

Please see our full paper here.

1. Setup
2. Replicating Data

This project uses Python version 3.11, and the other package versions are in the requirements.txt.

Note that pytorch must be installed with CUDA 11.8 enabled for this project to run properly. The installation instructions for pytorch can be found here.

Make sure that the dependencies are installed with administrator priveleges to avoid a permissions conflict.

# If you are using conda, and want to create a conda environment run this command:
conda env create -f environment.yaml python=3.11

# Alternatively, to install dependencies with pip, then run the following:
pip install -r requirements.txt

If you want to replicate the experiments in the paper, utilize either LitTraining.yaml (parameters used in an implementation of the work by Mnih et al.) or Tuned_Hyperparameters.yaml (parameters found by hyperparameter optimization with Optuna in this work). For Optuna hyperparameter tuning, 100 trials were conducted on 5000 episodes per trial.

# Run the following command to run the values used in the implementation
python Breakout.py -c ./LitTraining.yaml

# Run the following command to run the tuned hyperparameter values
python Breakout.py -c ./Tuned_Hyperparameters.yaml