Cognac Benchmark Example

Simple implementation of multi-agent reinforcement learning (MARL) algorithms for benchmarking purposes on 🥃COGNAC network environments.

COGNAC is a collection of MARL environment with network structure.

---

Features

• Lightweight: Minimal dependencies for quick setup.
• Modular: Easy-to-extend structure for adding new algorithms or environments.
• Benchmark-ready: Standardized scripts to evaluate and compare MARL performance.

---

Getting Started

Prerequisites

• Python 3.12 or above
• pip package manager

Installation

1. Clone the repository:

   git clone https://github.com/yojul/cognac-benchmark-example.git
   cd cognac-benchmark-example

2. Install dependencies:

   pip install -r requirements.txt

---

Usage

Running Experiments

All benchmark scripts are located in the algos/ directory. They are directly adapted from CleanRL implementations and coded as self-contained single file implementation of standard MARL algorithms. To run a standard experiment: All hyperparameters should be directly specified within the python file in the Args section.

python algos/benchmark_ippo.py

---

Project Structure

├── algos/                # MARL algorithm implementations and benchmark scripts
│   ├── benchmark_centralized_ppo.py  # Centralized Proximal Policy Optimization, used as a baselines.
│   ├── benchmark_idqn.py             # Simple Independent Deep Q-Learning
│   ├── benchmark_ippo.py             # Simple Independent Proximal Policy Optimization
│   ├── benchmark_ippo_continuous.py  # Simple Independent Proximal Policy Optimization handling continuous action space (use for Multi-commodity flow problem)
│   ├── benchmark_mappo.py            # Multi-Agent PPO algorithm introduced by (Yu & al., 2022)
│   └── benchmark_qmix.py             # Q-Mix algorithm introduced by (Rashid & al., 2018)
├── requirements.txt      # Python dependencies
├── LICENSE.txt           # Apache-2.0 License
└── README.md             # Project overview and instructions

---

Contributing

Contributions are welcome! To propose a new feature or algorithm:

---

License

This project is licensed under the Apache-2.0 License. See LICENSE.txt for details.

---

Contact

For questions or feedback, please open an issue or contact the maintainer:

• Repo Owner: Jules Sintes, INRIA Paris, DIENS, École Normale Supérieure, PSL University.
• Email: jules.sintes@inria.fr