corl 1D Docking Example

Intro

This document explains how to quickly get started working with the corl. It will walk you through installation of the corl repo and dependencies and how to launch a training loop for a simple environment: 1D Docking.

The code for the 1D Docking environment serves as a documented example of how to interface with the corl framework to create a custom environment.

Installation

First clone the corl git repository (https://git.act3-ace.com/act3-rl/corl) to your working directory. For example:

git clone https://git.act3-ace.com/act3-rl/corl.git

Navigate to the root of the local repository and use pip to install corl dependencies with the following commands. Note: setting up a project-specific environment is recommended before running these commands.

pip install -e path/to/corl
pip install -r path/to/corl/requirements.txt
pip install tensorflow

Training

To launch a training loop, the module /corl/train.py is used. This module must be passed the necessary config files at launch. From the root of the repository, execute the following command:

python corl/train_rl.py --cfg config/experiments/docking_1d.yml --compute-platform local

The config path after the --config flag defines the task, while the three strings following the -ac flag define the agent name, agent config path, and platform config path respectively. Multiple -ac flags may be used to define multiagent environments.