PyVRP is an open-source, state-of-the-art vehicle routing problem (VRP) solver. It currently supports VRPs with:
- Pickups and/or deliveries between depots and clients (capacitated VRP, VRP with simultaneous pickup and delivery, VRP with backhaul);
- Vehicles of different capacities, costs, and shift durations (heterogeneous fleet VRP);
- Time windows, client service durations, and release times (VRP with time windows and release times);
- Multiple depots (multi-depot VRP);
- Optional clients with prizes for visiting (prize collecting, team orienteering problem).
The implementation builds on Thibaut Vidal's HGS-CVRP, but has been completely redesigned to be easy to use as a highly customisable Python package, while maintaining speed and state-of-the-art performance. Users can customise various aspects of the algorithm using Python, including population management, crossover strategies, granular neighbourhoods and operator selection in the local search. Additionally, for advanced use cases such as supporting additional VRP variants, users can build and install PyVRP directly from the source code.
PyVRP is available on the Python package index as pyvrp.
It may be installed in the usual way as
pip install pyvrp
This also resolves the few core dependencies PyVRP has. The documentation is available here.
If you are new to vehicle routing or metaheuristics, you might benefit from first reading the introduction to VRP and introduction to HGS pages.
We provide some example notebooks that show how PyVRP may be used to solve vehicle routing problems. These include:
- A short tutorial and introduction to PyVRP's modelling interface, here. This is a great way to get started with PyVRP.
- A notebook solving classical VRP variants, here. In this notebook we solve several benchmark instances of the CVRP and VRPTW problems. We also demonstrate how to use the plotting tools available in PyVRP to visualise the instance and statistics collected during the search procedure.
- A notebook implementing a
solvemethod using PyVRP's components, here. This notebook is a great way to dive deeper into how PyVRP works internally.
We are very grateful for any contributions you are willing to make. Please have a look here to get started. If you aim to make a large change, it is helpful to discuss the change first in a new GitHub issue. Feel free to open one!
If you are looking for help, please follow the instructions here.
If you use PyVRP in your research, please consider citing the following paper:
Wouda, N.A., L. Lan, and W. Kool (2024). PyVRP: a high-performance VRP solver package. INFORMS Journal on Computing, forthcoming. https://doi.org/10.1287/ijoc.2023.0055
Or, using the following BibTeX entry:
@article{Wouda_Lan_Kool_PyVRP_2024,
doi = {10.1287/ijoc.2023.0055},
url = {https://doi.org/10.1287/ijoc.2023.0055},
year = {2024},
publisher = {INFORMS},
author = {Niels A. Wouda and Leon Lan and Wouter Kool},
title = {{PyVRP}: a high-performance {VRP} solver package},
journal = {INFORMS Journal on Computing},
}