If you have any issues during development, you can contact our team using github issues, or alternativelly through [email protected].
The full dataset is available to download here. Make sure you unzip the dataset into the data file. It should contain two directories named cvrp-instances-1.0
and delivery-instances-1.0
.
Alternativelly, you can generate the instances yourself from public data using the generation pipeline.
To correctly evaluate distances, you should use OpenStreetMaps distances provided by the OSRM server. Our recommended way of running OSRM is Docker. To run it, please follow these steps.
- Download and install docker, follow the instructions according to your operational system.
- Download our precompiled distance files (5.3Gb compressed, 12.6Gb decompressed).
- Extract the files into an
osrm
directory. - Run an OSRM backend container with the following command:
docker run --rm -t -id \
--name osrm \
-p 5000:5000 \
-v "${PWD}/osrm:/data" \
osrm/osrm-backend osrm-routed --algorithm ch /data/brazil-201110.osrm --max-table-size 10000
For more information, check our OSRM detailed documentation.
We provide an API for loading and running Python solvers. It currently supports any Python version >= 3.7.1, which is natively available in most up-to-date operating systems.
This project uses Python Poetry to manage dependencies. You can follow its docs to install it, but a simple
pip install poetry
# Or with sudo to install it system-wide
# sudo pip install poetry
normally suffices. Check if it worked with poetry --version
.
Then, at the root of the project install the dependencies with
poetry install
With everything in place, any Python command can be executed by preceding it with poetry run
(e.g., poetry run pytest tests/
). This is usually enough for executing the code in this project, but the user who demands more information can check the Poetry's website.
To implement a new method, we suggest you to create a Python solve
function that takes an instance and outputs the solution to a file.
from loggibud.v1.types import CVRPInstance, CVRPSolution
# Implement your method using a solve function that takes an instance and returns a solution.
def solve(instance: CVRPInstance) -> CVRPSolution:
return CVRPSolution(...)
# Loading an instance from file.
instance = CVRPInstance.from_file("path/to/instance.json")
# Call your method specific code.
solution = solve(instance)
# Saving your solution to a file.
solution.to_file("path/to/solution.json")
To evaluate your solution inside Python, you can do:
from loggibud.v1.eval.task1 import evaluate_solution
distance_km = evaluate_solution(instance, solution)
If you don't use Python, you should implement your own IO functions. The JSON schemas for reading and writing solutions are described below.
CVRPInstance
{
// Name of the specific instance.
"name": "rj-0-cvrp-0",
// Hub coordinates, where the vehicles originate.
"origin": {
"lng": -42.0,
"lat": -23.0
},
// The capacity (sum of sizes) of every vehicle.
"vehicle_capacity": 120,
// The deliveries that should be routed.
"deliveries": [
{
// Unique delivery id.
"id": "4943245fb66541edaf54f4e3aaed188a",
// Delivery destination coordinates.
"point": {
"lng": -43.12589115884953,
"lat": -22.89585186478512
},
// Size of the delivery.
"size": 2
}
// ...
]
}
CVRPSolution
{
// Name of the specific instance.
"name": "rj-0-cvrp-0",
// Solution vehicles.
"vehicles": [
{
// Vehicle origin (should be the same on CVRP solutions).
"origin": {
"lng": -43.374124642209765,
"lat": -22.790683484127058
},
// List of deliveries in the vehicle.
"deliveries": [
{
"id": "54b10d6d-2ef7-4a69-a9f7-e454f81cdfd2",
"point": {
"lng": -43.44893966650845,
"lat": -22.742762573031424
},
"size": 8
}
// ...
]
}
// ...
]
}
poetry run python -m loggibud.v1.eval.task1 \
--instance tests/results/cvrp-instances/train/rj-0-cvrp-0.json \
--solution results/rj-0-cvrp-0.json