keangkangkang / allset

This is the GitHub repository for our ICLR22 paper: "You are AllSet: A Multiset Function Framework for Hypergraph Neural Networks"

AllSet

This is the repo for our paper: You are AllSet: A Multiset Function Framework for Hypergraph Neural Networks. We prepared all codes and a subset of datasets used in our experiments.

All codes and script are in the folder src, and a subset of raw data are provided in folder data. To run the experiments, please go the the src folder first.

Enviroment requirement:

This repo is tested with the following enviroment, higher version of torch PyG may also be compatible.

First let's setup a conda enviroment

conda create -n "AllSet" python=3.7
conda activate AllSet

Then install pytorch and PyG packages with specific version.

conda install pytorch==1.4.0 torchvision==0.5.0 cudatoolkit=10.0 -c pytorch
pip install torch-scatter==2.0.4 -f https://pytorch-geometric.com/whl/torch-1.4.0+cu100.html
pip install torch-sparse==0.6.0 -f https://pytorch-geometric.com/whl/torch-1.4.0+cu100.html
pip install torch-cluster==1.5.2 -f https://pytorch-geometric.com/whl/torch-1.4.0+cu100.html
pip install torch-geometric==1.6.3 -f https://pytorch-geometric.com/whl/torch-1.4.0+cu100.html

Finally, install some relative packages

pip install ipdb
pip install tqdm
pip install scipy
pip install matplotlib

Generate dataset from raw data.

To generate PyG or DGL dataset for training, please create the following three folders:

p2root: './data/pyg_data/hypergraph_dataset_updated/'
p2raw: './data/AllSet_all_raw_data/'
p2dgl_data: './data/dgl_data_raw/'

And then unzip the raw data zip file into p2raw.

Run one single experiment with one model with specified lr and wd:

source run_one_model.sh [dataset] [method] [MLP_hidden_dim] [Classifier_hidden_dim] [feature noise level]

Note that for HAN, please check the readme file in ./src/DGL_HAN/.

To reproduce the results in Table 2(with the processed raw data)

source run_all_experiments.sh [method]

Notably, if you just want to reproduce the performance of AllSetTransformer in Table 2 without hyperparameter tuning, you can just run:

source run_AllSetTransformer.sh

Remark: We do not fix the random seed in our code so the results might be slightly different. If you find a huge discrepancy, please open an issue.

Compatibility with higher version of PyG

Please check the Issue #1 for more details. We still recommend to run our code with the exactly same version of PyG.

Issues

If you have any problem about our code, please open an issue and @ us (or send us an email) in case the notification doesn't work. Our email can be found in the paper.

Citation

If you use our code or data in your work, please cite our paper:

@inproceedings{
chien2022you,
title={You are AllSet: A Multiset Function Framework for Hypergraph Neural Networks},
author={Eli Chien and Chao Pan and Jianhao Peng and Olgica Milenkovic},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=hpBTIv2uy_E}
}