RGCN

RGCN[ESWC 2018]

• paper: Modeling Relational Data with Graph Convolutional Networks
• Author's code for entity classification: https://github.com/tkipf/relational-gcn
• Author's code for link prediction: https://github.com/MichSchli/RelationPrediction
• Code from dgl Team: rgcn / rgcn-hetero.
• Note: The difference between rgcn and rgcn-hetero is the storage form of node/edge type. The former use the dgl.NID stored in g.ndata, but the latter use the HeteroNodeView/HeteroEdgeView. And we use the latter storage form and nearly all models in OpenHGNN use the latter storage form as possible.

How to run

• Clone the Openhgnn-DGL

  # For node classification task
  python main.py -m RGCN -t node_classification -d aifb -g 0 --use_best_config
  # For link prediction task
  python main.py -m RGCN -t link_prediction -d HGBl-amazon -g 0 --use_best_config

  If you do not have gpu, set -gpu -1.

  Supported dataset

  Node classification:

  RDFDataset[aifb/mutag/bgs/am], HGBn and other datasets for node classification.

  Link prediction:

Performance

Task: Node classification

Evaluation metric: accuracy

Method	AIFB	MUTAG	BGS	AM
RGCN	97.22	72.06	96.55	88.89

Task: Link prediction

Evaluation metric: roc_auc

Method	AIFB	MUTAG	BGS	AM
RGCN(dgl)（best)	97.22	72.06	96.55	88.89

TrainerFlow: node classification flow

Model

• dglnn.HeteroGraphConv
• dglnn.WeightBasis [The basis function decomposition]

Hyper-parameter specific to the model

You can modify the parameters[RGCN] in openhgnn/config.ini.

Best config can be found in best_config.

More

Contirbutor

Tianyu Zhao[GAMMA LAB]

If you have any questions,

Submit an issue or email to [email protected].