Mixed-Integer Linear Optimization via Learning-Based Two-Layer Large Neighborhood Search

This repository is an implementation of the LION19 paper: Mixed-Integer Linear Optimization via Learning-Based Two-Layer Large Neighborhood Search.

Software dependencies

python=3.11
pytorch=2.1.0
scip=8.0.4
gurobi=10.0.3
torch-geometric=2.4.0
pytorch-metric-learning=2.5.0

Running experiments

Use the following command to generate training/testing instances:

python instance_generate.py --usage 'test' --instance 'SC' --number 10
python instance_generate.py --usage 'train' --instance 'SC' --number 1000

For training data collection:

python data_collection.py

For training model:

python train.py

You can directly utilize the trained model in this repo, and for inference, use the following command:

bash run.sh

For the comparison:

python result.py

Citing our work

If you would like to utilize TLNS in your research, we kindly request that you cite our paper as follows:

@article{liu2024mixed,
  title={Mixed-Integer Linear Optimization via Learning-Based Two-Layer Large Neighborhood Search},
  author={Liu, Wenbo and Wang, Akang and Yang, Wenguo and Shi, Qingjiang},
  journal={arXiv preprint arXiv:2412.08206},
  year={2024}
}