BIPNN: Learning to Solve Binary Integer Programming via Hypergraph Neural Networks

This repository is the official implementation of BIPNN: Learning to Solve Binary Integer Programming via Hypergraph Neural Networks

Requirements

To install requirements, you need anaconda or miniconda and run:

conda env create -f environment.yaml

Run code

We provide a beautiful notebook file example/desc.ipynb that combines code to restate our work in the paper and reproduces some experimental results for your reference, You can either just browse it using jupyter or try to re-run it even if it already contains all the results.

In addition, we have provided some experiments in the test/ folder

Project directory structure

📦src
 ┣ 📂maxcut
 ┃ ┣ 📜__init__.py
 ┃ ┣ 📜loss.py
 ┃ ┗ 📜utils.py
 ┣ 📜__init__.py
 ┣ 📜constraints.py
 ┣ 📜core.py
 ┣ 📜loss.py
 ┣ 📜models.py
 ┣ 📜train.py
 ┗ 📜utils.py

Results

The Effectiveness of BIPNN’s Hypergraph Generation Mechanism:

Comparison with Linearization-based BIP Solvers for Hypergraph maxcut problems:

Conversion algorithm

Annealing Strategy