This repo contains the companion code for our paper:

HNHN: Hypergraph Networks with Hyperedge Neurons
By Yihe Dong, Will Sawin, and Yoshua Bengio

Presented at the Graph Representation Learning and Beyond workshop at ICML 2020.

One can run training on the main HNHN script as follows:

E.g. specify dataset name, set random seed, and number of layers:

python hypergraph.py --dataset_name citeseer --seed --n_layers 1

E.g. use a dimension-reduced version of the dataset:

python hypergraph.py --dataset_name cora --do_svd

The functions that call on baseline repository codes are in baseline.py, and can be called as follows:

python baseline.py --dataset_name citeseer --method hgnn

For all runtime options, please see either utils.py or run:

python hypergraph.py --h

Note: baselines.py assumes the presence of two baseline repos: HyperGCN and HGNN. If one wishes to run these baseline models, their directories should be named as hypergcn and hgnn, respectively, and should be placed in the parent directory of the current repo, as indicated in the _init_paths.py script.

data.py extracts and processes raw data as described in the paper. For instance the Cora Information Extraction data.

For an example of processed data please see the CiteSeer data (citeseer.pt contains the same hypergraph but with reduced feature dimension).

p.s. please tag me when creating an issue, GitHub currrently cannot notify me of new issues.

If you find our work helpful, please cite as:

@article{HNHN2020,
  title         = {HNHN: Hypergraph Networks with Hyperedge Neurons},
  author 	= {Dong, Yihe and Sawin, Will and Bengio, Yoshua},
  url       	= {https://arxiv.org/abs/2006.12278},
  journal 	= {ICML Graph Representation Learning and Beyond Workshop},
  year          = {2020}
  }