jiqingjiang / graph-learning Goto Github PK

图模型项目（GCN、GAT、GraphSAGE、deepwalk、node2vec）细节实践、论文复现、持续更新、欢迎star、交流学习。

based on dgl and pytorch mainly

pip install -r requirements.txt

download dataset，put it to ./data/

uploaded dataset blog already

Notes of model written here:

1. 游走图模型--同构图DeepWalk解析
2. 游走图模型-聊聊Node2Vec
3. 图卷积：从GCN到GAT、GraphSAGE
4. 怎么搭一个GCN？只需这四步
5. 怎么搭好一个GraphSAGE？按这三步走
6. Link-Prediction：搭一个无监督的GraphSAGE

①. How to run deepwalk model for graph embedding？

cd deepwalk
python main.py

②. node classification task

python node_classification.py

①. How to run Node2Vec model

cd node2vec
python main.py

②. node classification task(should chang the checkpoint of node2vec in node_classification.py).

python node_classification.py

①. How to run GCN model

python train.py

Cora dataset node classification(cora dataset will be download in ~/.dgl/ automatically).
Test accuracy ~0.806 (0.793-0.819) (paper: 0.815).

①. How to run GraphSAGE model

cd graphsage/node_classification
python train.py

Cora dataset node classification(cora dataset will be download in ~/.dgl/ automatically).
Test accuracy ~0.781(0.762-0.801) (paper: 0.815).

①. How to run GraphSAGE model

cd graphsage/link_prediction
python train.py

Test F1: 0.630 (0.6120.648) (cora数据集)

①. How to run GAT model

python train.py

Cora dataset node classification(cora dataset will be download in ~/.dgl/ automatically).
Test accuracy ~0.810 (0.792-0.820) (paper: 0.830).