This repo reimplements graph neural network architectures.
python main.py --dataset {dataset_name}
for example,
python main.py --dataset cora
For now, most of the dataset pipeline is from: https://github.com/tkipf/gcn
- citeseer [explaination] [rawdata]
- cora [explaination] [rawdata]
- pubmed [explaination] [rawdata]
- NELL (To be updated)
- GCN: Kipf, T. N., & Welling, M. (2017). Semi-Supervised Classification with Graph Convolutional Networks. International Conference on Learning Representations (ICLR).
Numbers in parenthesis states results from original paper.
Method | Citeseer | Cora | Pubmed | NELL |
---|---|---|---|---|
GCN [Kipf & Welling] | 68.9 (70.3) | 80.79 (81.5) | 78.9 (79.0) |
GCN(Kipf)
- (Citeseer) 100 epochs, dropout rate = 0.5, L2 regularization with 5e-04, num_hiddens of 16
- (Cora) 100 epochs, dropout rate = 0.5, L2 regularization with 5e-04, num_hiddens of 16
- (Pubmed) 100 epochs, dropout rate = 0.5, L2 regularization with 5e-04, num_hiddens of 16