PyTorch and TensorFlow2 implementation of Snowball and Truncated Krylov Graph Convolutional Network (GCN) architectures for semi-supervised classification [1].
This repository contains the Cora, CiteSeer and PubMed dataset.
Results are collected through the PyTorch implementation, which are published in our NeurIPS paper.
There are slight differences between the 2 implementations, so you may have to redo the hyperparameter search for the TensorFlow2 implementation.
Please feel free to leave comments if you have trouble reproducing the results!
- PyTorch 1.3.x or TensorFlow 2.x.x
- Python 3.6+
- Best with NVIDIA apex (we have used the NGC container with singularity)
python initialize_dataset.py
python train.py
[1] Luan, et al., Break the Ceiling: Stronger Multi-scale Deep Graph Convolutional Networks, 2019
Please kindly cite our work if necessary:
@incollection{luan2019break,
title = {Break the Ceiling: Stronger Multi-scale Deep Graph Convolutional Networks},
author = {Luan, Sitao and Zhao, Mingde and Chang, Xiao-Wen and Precup, Doina},
booktitle = {Advances in Neural Information Processing Systems 32},
editor = {H. Wallach and H. Larochelle and A. Beygelzimer and F. d\textquotesingle Alch\'{e}-Buc and E. Fox and R. Garnett},
pages = {10943-10953},
year = {2019},
publisher = {Curran Associates, Inc.},
url = {https://arxiv.org/abs/1906.02174}
}