awslabs/aws-last-mile-district-opt

AWS District Optimization for Last Mile Planning

This repository demonstrates our solution to the district optimization problem for package allocation and last mile planning. In our approach, we assemble basic geographic units, such as streets, post code areas, suburbs, into larger geographic districts known as polygons, in order to achieve some optimisation objectives or planning criteria. Packages are automatically allocated to different polygons based on the spatial matching between their addresses and the geo-fence of the established polygons. Our solution runs on SageMaker notebook instances.

Please check out our blog post Aramex Australia makes delivery easier for drivers using open source solutions from Amazon for technical details.

Installation

# Create the conda virtual environment
conda create --name pa-opt python=3.8
conda activate pa-opt
pip install -r requirements.txt

Quick Start

Going through Step 1 to Step 7 by running the following terminal commands on a SageMaker notebook instance. Step 2 and Step 3 can also be integrated as a SageMaker processing job.

# download data and prepare the basic geo-units
python 01_data_download_bgu.py

# follow instructions on 02_launch_vahalla.md to setup Vahalla API

# generate distance matrix
python 03_gen_dist_matrix.py

# generate the graph
python 04_gen_graph.py

# perform graph partitioning
python 05_graph_partition.py

# produce data for evaluation
python 06_eval_prep.py

# evaluation
python 07_evaluate.py

License

This project is licensed under the Apache-2.0 License.

Acknowledgements

• https://github.com/kw/metis-python (MIT License)
• https://github.com/KarypisLab/METIS (Apache-2.0 License)
• https://github.com/google/or-tools (Apache-2.0 License)
• https://github.com/valhalla/valhalla (MIT License)