Paper: Heterogeneous Graph Attention Network
Code from author: https://github.com/Jhy1993/HAN
Code from dgl Team: https://github.com/dmlc/dgl/tree/master/examples/pytorch/han
We use the code from dgl Team and integrate the model into our package.
Clone the Openhgnn-DGL
python main.py -m HAN -t node_classification -d acm_han_raw -g 0
If you do not have gpu, set -gpu -1.
Candidate dataset: acm_han_raw, acm4GTN, imdb4GTN
acm_han_raw | Macro-F1 | Micro-F1 |
---|---|---|
HAN[OpenHGNN] | 90.11 | 90.02 |
Node classification | acm4GTN | imdb4GTN |
---|---|---|
GTN[OpenHGNN] | 92.22 | 61.58 |
HAN[OpenHGNN] | 91.63 | 57.30 |
The model is trained in semi-supervisied node classification.
We implement HANLayer with MetapathConv.
SemanticAttention
Supported dataset: acm_han_raw, acm4GTN, imdb4GTN
You can download the dataset by
wget https://s3.cn-north-1.amazonaws.com.cn/dgl-data/dataset/acm_han_raw.zip
wget https://s3.cn-north-1.amazonaws.com.cn/dgl-data/dataset/acm4GTN.zip
wget https://s3.cn-north-1.amazonaws.com.cn/dgl-data/dataset/imdb4GTN.zip
num_channels = 2 # number of channel
num_layers = 3 # number of layer
adaptive_lr_flag = True # use different learning rate for weight in GTLayer.
Best config can be found in best_config
Tianyu Zhao[GAMMA LAB]
Submit an issue or email to [email protected].