DiffILO: Differentiable Integer Linear Programmin

This repository contains the implementation of DiffILO, an unsupervised learning approach for predicting solutions to Integer Linear Programs (ILPs).

Paper: https://openreview.net/pdf?id=FPfCUJTsCn

Note: This is the latest version, and we're still in the process of organizing and refining the code. Updates will follow. Feel free to reach out with any questions.

Environment Setup

• Python environment

  • python 3.8
  • pytorch 2.3.0
  • torch-geometric 2.6
  • ecole 0.8.1
  • pyscipopt 4.4.0
  • gurobipy 10.0
  • tensorboardX

• MILP Solver

  • Gurobi 11.0.3. Academic License.

• Hydra

  • Hydra for managing hyperparameters and experiments.

To set up the environment, run the installation script:

bash scripts/environment.sh

Alternatively, create the environment from a file:

conda env create -f scripts/environment.yml

Project Structure

The workspace is organized as follows:

DiffILO
├── conf/               # Hydra config files
├── data/               # Dataset directory (see below)
│   ├── CA/
│   │   ├── train/
│   │   └── test/
│   ├── IS/
│   │   ├── train/
│   │   └── test/
│   └── SC/
│       ├── train/
│       └── test/
├── scripts/            # Training/testing scripts
├── src/                # Core implementation
├── preprocess.py       # Data preprocessing
├── train.py            # Training entry point
├── test.py             # Evaluation entry point
└── README.md

Data Format

The dataset is located in the data/ directory. Data instances can be in any format, as long as they are compatible with scip and ecole. An example dataset is available at this link.

Usage

1. Preprocessing

To preprocess a dataset (e.g., SC), run:

python preprocess.py dataset=SC num_workers=50

2. Training DiffILO

To train the model with default settings:

python train.py dataset=SC cuda=0

3. Test DiffILO

To evaluate the trained model,

python test.py dataset=SC cuda=0

Citation

If you find DiffILO useful or relevant to your research, please consider citing our paper.

@inproceedings{
    geng2025differentiable,
    title={Differentiable Integer Linear Programming},
    author={Zijie Geng and Jie Wang and Xijun Li and Fangzhou Zhu and Jianye HAO and Bin Li and Feng Wu},
    booktitle={The Thirteenth International Conference on Learning Representations},
    year={2025},
    url={https://openreview.net/forum?id=FPfCUJTsCn}
}