PyMARL Deploy

This repository contains scripts to deploy the Python Multi-Agent Reinforcement Learning (PyMARL) framework to Linux based systems. PyMARL execution environments can be setup using either Anaconda or Docker

Configurations

Deployment configurations for Linux can be found in the config directory and can be run on the command line e.g.:

bash config/nvidia-docker-ubuntu.sh <REPO> <INSTALL_DIR>

will install NVIDIA driver dependencies as well as Docker and NVIDIA Docker on Ubuntu. The first argument is the target PyMARL Git repository to clone, the second argument is the location to download this repository to e.g.:

bash config/nvidia-docker-ubuntu.sh git@github.com:sina-masoud-ansari/pymarl.git /data/pymarl

Similarly:

bash config/nvidia-conda-ubuntu.sh <REPO> <INSTALL_DIR>

will install the NVIDIA pre-requisites before setting up an Anaconda environment named pymarl which should then be activated before running PyMARL.

Note: This assumes a prexisting Anaconda environment. One can be installed from the Anaconda webiste. Miniconda is a suitable light-weight alternative.

If NVIDIA drivers are already installed, one can instead run:

bash config/docker-ubuntu.sh <REPO> <INSTALL_DIR>

or similar to avoid installing the drivers again.

Note: The NVIDIA dependencies are currently setup for Ubuntu 18.04. You should be able to change the nvidia config to suit your target system. Note that nvidia-docker is not supported on all Ubuntu releases.

Repositories

The deployment scripts are have been tested with the https://github.com/sina-masoud-ansari/pymarl fork as some changes were required in the original Dockerfile and requirements.txt for PyMARL to run correctly.

Recommended Environments

Amazon Web Services

The following images are recommended on p2.* and p3.* instances

• Deep Learning Base AMI (Ubuntu) with config/*-ubuntu.sh
• Deep Learning AMI (Ubuntu) with config/*-ubuntu.sh
• Ubuntu Server 18.04 LTS with config/nvidia-*-ubuntu.sh