THU-HyperG: A python toolbox for hypergraph learning

Introduction

THU-HyperG is a python toolbox for hypergraph learning. Hypergraph is a generalization of graph, which is composed of a set of nodes and a set of hyperedges. Different from simple graph, in which each edge connect a pair of nodes, each hyperedge can connect any number of nodes in hypergraph. The flexible edge degree in hypergraph enables the hypergraph model to formulate the high-order correlation of data.

Hypergraph learning is mainly consisted of two procedures, hypergraph generation and learning on hypergraph. Thus, in this toolbox, we provide several hypergraph generation methods and learning methods on hypergraph. Some useful utilities are also included in this toolbox. Besides, we provide some examples about how to apply this toolbox to the tasks like classification and segmentation.

Specifically, it provides the following functionalities:

• Hypergraph generation methods: knn based, epsilon ball based, clustering based, l1 representation based and grid based methods.
• Learning on hypergraph methods: transductive learning, inductive learning, diffusion and clustering methods.
• Utils: some preprocessing and evaluation methods.

Installation

You can install from source

git clone https://github.com/iMoonLab/THU-HyperG.git
cd THU-HyperG
pip install .

Examples

We provide some examples in the examples/ directory about how to apply this toolbox to the tasks like classification and segmentation.

Citing

if you find our work useful in your research, please consider citing:

@article{gao2020hypergraph,
  title={Hypergraph Learning: Methods and Practices.},
  author={Gao, Y and Zhang, Z and Lin, H and Zhao, X and Du, S and Zou, C},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2020}
}

Contributing

We appreciate all contributions to improve THU-HyperG. Pull requests are always welcomed.