ORSuite: Benchmarking Suite for Sequential Operations Models
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.