Inductive-Hypergraph-Neural-Network-based-on-Hypergraph-Sampler

The repo implement a Inductive Hypergraph Neural Network(Inductive-HGNN) with Hypergraph Neighbors Sampler(HySampler), and perform this on OGB-products.

We implement Hypergraph Neighbors Sampler by C++ instead of Python to improve the performance, there are almost 50 times improvement, which is necessary for train in large scale dataset, e.g. ogb-products.

Although the Inductive-HGNN do not have a comparable performance for ogb-products(maybe some improvements can be made for HGNN), we think the faster HySampler is userful for the researchers who are interested in HGNN.

Unfortunately I ended the internship in iMoon-lab and didn't have computing resources(GPUs) to do more following works. If you are interested in this work, please contact my Email(djytyang@Gmail.com) for cooperation.

Installation and Run

Dependency

• PyTorch
• PyTorch-Geometric
• THU-DeepHypergraph
• tqdm

Install C++ version Hypergraph Neighbor Sampler

cd hysample_cpp
python setup.py install 

Run

python cora_main.py
python ogb_main.py