Pytorch 1.0 with python 3.7.
GCN implementation for paper: Semi-Supervised Classification with Graph Convolutional Networks
| dataset | Citeseea | Cora | Pubmed | NELL |
|---|---|---|---|---|
| GCN(official) | 70.3 | 81.5 | 79.0 | 66.0 |
| This repo. | 81.7 | 78.7 |
Hints:
To achieve 81.7% on Cora dataset, pytorch need 400 epochs.
and it train relatively slower on Pubmed dataset.
python train.py
