This is an unofficial Tensorflow implementation of the paper: DropEdge: Towards Deep Graph Convolutional Networks on Node Classification. I simply add the random sampling part to the orignal GCN.
- Python 3.6.2
- tensorflow (>0.12)
- networkx
*python train.py --dataset cora
* percent: sampling percent
* normalization: normalization of adjacency matrix.
* task_type: semi or full. semi uses 120 nodes for trainning in citeseer and 140 nodes in cora.
@inproceedings{
rong2020dropedge,
title={DropEdge: Towards Deep Graph Convolutional Networks on Node Classification},
author={Yu Rong and Wenbing Huang and Tingyang Xu and Junzhou Huang},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=Hkx1qkrKPr}
}
@inproceedings{kipf2017semi,
title={Semi-Supervised Classification with Graph Convolutional Networks},
author={Kipf, Thomas N. and Welling, Max},
booktitle={International Conference on Learning Representations (ICLR)},
year={2017}
}