Graph neural networks (GNN) are drawing more and more attention and achieving remarkable improvement in various domains. Inspired by impressive recent paper benchmarking different GNNs, this repository aims to catch up many existing GNN variants without complicated codes and experiments.
Codes are implemented with PyTorch and DGL. For DGL, you can find kind tutorial on the official website. Codes refer to many part of codes from DGL tutorial (especially GCN, GAT).
Cora Citation Network dataset is a transductive graph dataset with scientific papers as nodes and citation relationship as edges. It contains 2.7K nodes and 5.4K edges. Given a citation graph, the task is to classify each node to category it belongs to. Total number of category is 7.
The following models are currently implemented. More to be added.
- Graph Convolution Networks (Kipf & Welling, Semi-Supervised Classification with Graph Convolutional Networks)
- Graph Attention Networks (Veličković et al., Graph Attention Networks)
run.py is all you need to run. All arguments for experiment settings and GNNs are set with argparse package and they can be found in utils/config.py.
For example, if you want to run 2-layer GCN Classifier with hidden_dim=50, run the script below:
python run.py --gnn GCN --hidden_dim 50 --num_layers 2
Other parameters, such as learning rate or # of epochs, are set as default values.
HAVE FUN !
Table below shows experimental results from the original paper and the current repo on Cora dataset (without exhaustive parameter search).
| Model | Accuracy |
|---|---|
| GCN (paper) | 81.5 |
| GCN (repo) | 80.6 |
| GAT (paper) | 83.0 |
| GAT (repo) | 82.0 |