Light-MILPopt: Solving Large-scale Mixed Integer Linear Programs with Lightweight Optimizer and Small-scale Training Dataset

Contact

We welcome academic and business collaborations with funding support. For more details, please contact us via email at xuhua@tsinghua.edu.cn.

Overview

This release contains the core processes of the Light-MILPopt framework, as described in the paper Light-MILPopt: Solving Large-scale Mixed Integer Linear Programs with Lightweight Optimizer and Small-scale Training Dataset. The provided code implements the main components of the approach, covering data generation, training, and inference. Interfaces are provided for flexibility, allowing easy adaptation to different contexts.

Here is a brief overview of the provided files. More detailed documentation is available within each file:

• Code/data_generation.py: Generates standard mixed integer linear programming (MILP) problems for training and testing datasets.
• Code/data_solution.py: Processes training data by generating optimal solutions for MILP problems, which are used as training labels.
• Code/EGAT_layers.py: Defines the layer structure of the EGAT (Enhanced Graph Attention Network) model.
• Code/EGAT_models.py: Defines the overall structure and architecture of the EGAT model.
• Code/train.py: Contains the training process for the EGAT model.
• Code/test.py: Carries out the testing process to evaluate the performance of the model on test datasets.

Requirements

The required environment is specified in the provided YAML file (e.g., Code/Light_MILPopt.yml).

Usage

The code should be run in the following order to perform training and testing:

1. data_generation.py: Generate training and testing datasets.
2. data_solution.py: Generate optimal solutions for training datasets.
3. EGAT_layers.py: Define the layers of the EGAT model.
4. EGAT_models.py: Define the overall EGAT model architecture.
5. train.py: Train the EGAT model using the generated datasets.
6. test.py: Run the trained model on test datasets to obtain optimized results.

Citing this work

If you use the code or methodology from this repository, please cite the following paper:

@inproceedings{ye2023light,
  title={Light-MILPopt: Solving Large-scale Mixed Integer Linear Programs with Lightweight Optimizer and Small-scale Training Dataset},
  author={Ye, Huigen and Xu, Hua and Wang, Hongyan},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2023}
}