Inducing Cooperation through Reward Reshaping based on Peer Evaluations in Deep Multi-Agent Reinforcement Learning

Primary author: David Earl Hostallero (david.hostallero@mail.mcgill.ca)

Overview

This is a reimplementation of the algorithms in the paper. The gems directory contains the simulator and algorithm source codes for the Resource Sharing environment. The pursuit directory contrains the simulator and algorithm source codes for the Paritally-Cooperative Pursuit.

Running the program

Resource Sharing

cd gems
python main.py --folder=gems --seed=1

Partially-Coopertive Pursuit

cd pursuit
pythona main.py --folder=pursuit --seed=1

Parameters

• --seed: the seed number for pseudo-random number generation
• --folder: subfolder where you want to save the logfiles and weights

Disclaimer

This repository uses Python 2 and TensorFlow 1. Newer versions of the Python and TF may not be able to support this.