ORSuite: Benchmarking Suite for Sequential Operations Models
Due to recent updates to the Gymnasium API, including changes in environment conventions and notation, the current ORSuite package is temporarily incompatible with the latest Gymnasium releases. We are actively updating the package to restore compatibility.
For questions or updates, please contact sean.sinclair@northwestern.edu.
Reinforcement learning (RL) is a natural model for problems involving real-time sequential decision making. In these models, a principal interacts with a system having stochastic transitions and rewards and aims to control the system online (by exploring available actions using real-time feedback) or offline (by exploiting known properties of the system).
These project revolves around providing a unified landscape on scaling reinforcement learning algorithms to operations research domains.
https://orsuite.readthedocs.io/en/latest/
https://colab.research.google.com/drive/1oSv8pCwl9efqU4VEHgi8KXNvHiPXi7r1?usp=sharing
In order to install the required dpeendencies for a new conda environment, please run:
conda create --name ORSuite python=3.8.5
conda activate ORSuite
pip install -e .
The repository has three main components as a traditional Reinforcement Learning set-up :
- Environments : Environment for the agent to interact with and reside in.
~/or_suite/envs - Agents : Choice of Algorithm
~/or_suite/agents - Experiments : This is a take on implementing the enviroment and agents with a choice of algorithm
~/or_suite/experiment
See 'ORSuite Contribution Guide' to see information on how to add new environments and algorithms to the package.