huyhoang17 / GCNII

PyTorch implementation of "Simple and Deep Graph Convolutional Networks"

Simple and Deep Graph Convolutional Networks

This repository contains a PyTorch implementation of "Simple and Deep Graph Convolutional Networks".(https://arxiv.org/abs/2007.02133)

Dependencies

• CUDA 10.1
• python 3.6.9
• pytorch 1.3.1
• networkx 2.1
• scikit-learn

Datasets

The data folder contains three benchmark datasets(Cora, Citeseer, Pubmed), and the newdata folder contains four datasets(Chameleon, Cornell, Texas, Wisconsin) from Geom-GCN. We use the same semi-supervised setting as GCN and the same full-supervised setting as Geom-GCN. PPI can be downloaded from GraphSAGE.

Results

Testing accuracy summarized below.

Dataset	Depth	Metric	Dataset	Depth	Metric
Cora	64	85.5	Cham	8	62.48
Cite	32	73.4	Corn	16	76.49
Pubm	16	80.3	Texa	32	77.84
Cora(full)	64	88.49	Wisc	16	81.57
Cite(full)	64	77.13	PPI	9	99.56
Pubm(full)	64	90.30	obgn-arxiv	16	72.74

Usage

• To replicate the semi-supervised results, run the following script

sh semi.sh

• To replicate the full-supervised results, run the following script

sh full.sh

• To replicate the inductive results of PPI, run the following script

sh ppi.sh

Reference implementation

The PyG folder includes a simple PyTorch Geometric implementation of GCNII. Requirements: torch-geometric >= 1.5.0 and ogb >= 1.2.0.

• Running examples

python cora.py
python arxiv.py

Citation

@article{chenWHDL2020gcnii,
  title = {Simple and Deep Graph Convolutional Networks},
  author = {Ming Chen, Zhewei Wei and Zengfeng Huang, Bolin Ding and Yaliang Li},
  year = {2020},
  booktitle = {Proceedings of the 37th International Conference on Machine Learning},
}