CogDL implementation of VRGCN for ogbn-arxiv:
Jianfei Chen, Jun Zhu, Le Song. Stochastic Training of Graph Convolutional Networks with Variance Reduction. Paper in arXiv. In ICML'2018.
Requires CogDL 0.5-alpha0 or later versions.
# Run with model with default config
python main.py
For more hyper-parameters, please find them in the main.py.
Here are the results over 10 runs which are comparable with OGB official results reported in the leaderboard.
| Method | Test Accuracy | Validation Accuracy | #Parameters |
|---|---|---|---|
| VRGCN | 0.7224 ± 0.0042 | 0.7260 ± 0.0030 | 44,328 |