BO_UrbanMobility_Benchmark

This repository contains the code and resources for the workshop talk on Bayesian Optimization for High-dimensional Urban Mobility Problems presented at the NeurIPS 2024 Workshop on Bayesian Decision-making and Uncertainty. The project addresses optimization challenges in designing urban mobility digital twins and explores open questions and research opportunities in Bayesian optimization, uncertainty quantification, and inverse optimization.

Overview

The repository presents a framework for tackling travel demand (origin-destination) estimation in urban networks, specifically targeting two road networks:

1. A simplified illustrative toy network.
2. The San Francisco metropolitan network.

The codebase provides tools for:

• Setting up travel demand estimation problems for urban mobility applications.
• Applying Bayesian optimization techniques to solve high-dimensional optimization problems relevant to urban transportation systems.
• Evaluating optimization results with focus on uncertainty quantification and inverse optimization.

Repository Structure

• bayesian_optimization/: Contains Jupyter notebooks to run experiments and helper functions for Bayesian optimization tasks.
• config/: Configuration files specifying parameters for different networks and model settings.
• network/: Network data for the toy network and the San Francisco metropolitan network, including traffic and topology information.
• output/: Directory that will be automatically generated to store experiment results and logs.
• README.md: Project documentation.

Network Data

The data for the San Francisco metropolitan network and toy network can be downloaded from the following link: Google Drive - Network Data

After downloading, place the data in the network/ directory to ensure proper setup.

• network
  • quickstart
  • SFO

Getting Started

Prerequisites

Ensure you have Python 3.x installed. Recommended packages are listed in requirements.txt. Install dependencies with:

pip install -r requirements.txt

Running the Code

To start with a simple example:

1. Run the Jupyter notebooks bayesian_optimization/bo_vanilla.ipynb for step-by-step guidance on solving the travel demand estimation problem for each network.

Future Development Plans

Citation

If you use this code or find it helpful, please cite:

@article{choispatiotemporal,
  title={Bayesian Optimization for High-dimensional Urban Mobility Problems},
  author={Choi, Seongjin and Rodriguez, Sergio and Osorio, Carolina},
  booktitle={NeurIPS workshop on Bayesian Decision-making and Uncertainty 2024}
}

If you use toy network, please cite:

@inproceedings{SUMO2018,
          title = {Microscopic Traffic Simulation using SUMO},
         author = {Pablo Alvarez Lopez and Michael Behrisch and Laura Bieker-Walz and Jakob Erdmann and Yun-Pang Fl{\"o}tter{\"o}d and Robert Hilbrich and Leonhard L{\"u}cken and Johannes Rummel and Peter Wagner and Evamarie Wie{\ss}ner},
      publisher = {IEEE},
      booktitle = {The 21st IEEE International Conference on Intelligent Transportation Systems},
           year = {2018},
        journal = {IEEE Intelligent Transportation Systems Conference (ITSC)},
       keywords = {traffic simulation, modelling, optimization},
            url = {https://elib.dlr.de/124092/}
 }

If you use SFO network, please cite:

@article{ambuhl2022traffic,
  title={Traffic Simulations for Boston, Lisbon, Los Angeles, Rio de Janeiro, San Francisco: SUMO simulation files for running 24h of car traffic},
  author={Amb{\"u}hl, Lukas and Menendez, Monica and Gonz{\'a}lez, Marta C},
  year={2022},
  publisher={ETH Zurich}
}

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For questions or feedback, please contact Seongjin Choi or Carolina Osorio.