No pitfalls

This pygcn implementation is the same as TensorFlow implementation in https://github.com/tkipf/gcn, fixing the differences of data splits, normalization, dropout in the official https://github.com/tkipf/pygcn. data splits in the pygcn/utils.py

Performance

• cora: 0.820 (paper: 0.815)
• citeseer: 0.707 (paper: 0.703)
• pubmed: 0.794 (paper: 0.790)

Usage

python train.py --dataset cora --early_stopping 10
python train.py --dataset citeseer --early_stopping 10
python train.py --dataset pubmed --early_stopping 20
early_stopping is suggested to be 10 for cora and citeseer, 20 for pubmed.

References

PyTorch implementation of Graph Convolutional Networks (GCNs) for semi-supervised classification [1].
[1] Kipf & Welling, Semi-Supervised Classification with Graph Convolutional Networks, 2016

Requirements

• PyTorch 0.4 or 0.5
• Python 2.7 or 3.6