- Paper link: Simplifying Graph Convolutional Networks
- Author's code repo: https://github.com/Tiiiger/SGC.
- PyTorch 0.4.1+
- requests
bash pip install torch requests
The folder contains an implementation of SGC (sgc.py
).
sgc_reddit.py
contains an example of training SGC on the reddit dataset.
Run with following (available dataset: "cora", "citeseer", "pubmed")
python3 sgc.py --dataset cora --gpu 0
python3 sgc.py --dataset citeseer --weight-decay 5e-5 --n-epochs 150 --bias --gpu 0
python3 sgc.py --dataset pubmed --weight-decay 5e-5 --bias --gpu 0
Run the following command to train on the reddit dataset.
python sgc_reddit.py --gpu 0
On NVIDIA V100
- cora: 0.819 (paper: 0.810), 0.0008s/epoch
- citeseer: 0.725 (paper: 0.719), 0.0008s/epoch
- pubmed: 0.788 (paper: 0.789), 0.0007s/epoch
- reddit: 0.947 (paper: 0.949), 0.6872s in total