Georgia Tech OMSCS CS7643: Final Project

Super Mario Bros - Deep Reinforcement Learning

Proximal Policy Optimization (PPO) Implementation

mario-dl-ppo.mp4

Setup

In a python 3.7+ environment, run:

conda create --name mario python=3.9
conda activate mario
pip install torch==1.10.1+cu102 -f https://download.pytorch.org/whl/cu102/torch_stable.html
pip install -r requirements.txt

Yaml Configuration

Inside the file configs/ppo.yaml there are few keys that change the way the application runs. First you must train a model so only train is set to true. Then once you have a model saved, you can set train to False, test to True, then add a key load_model with the path to your model found in checkpoints/PPO/. If you would like to generate plots you must add keys test_csv_path and train_csv_path pointing to the csv you generate from running your training and your test runs.

Run

conda activate mario
python main.py

Check nvidia memory usage

nvidia-smi -l --query-gpu=timestamp,temperature.gpu,memory.used,memory.total --format=csv

Credits

Adopted the GAE calculation from https://github.com/uvipen/Super-mario-bros-PPO-pytorch