This DGL example implements the GNN model proposed in the paper GeniePath: Graph Neural Networks with Adaptive Receptive Paths.

This example was implemented by Kay Liu during his SDE intern work at the AWS Shanghai AI Lab.

• Python 3.7.10
• PyTorch 1.8.1
• dgl 0.7.0
• scikit-learn 0.23.2

The datasets used for node classification are Pubmed citation network dataset (tranductive) and Protein-Protein Interaction dataset (inductive).

If want to train on Pubmed (transductive), run

python pubmed.py

If want to use a GPU, run

python pubmed.py --gpu 0

If want to train GeniePath-Lazy, run

python pubmed.py --lazy True

If want to train on PPI (inductive), run

python ppi.py

Dataset: Pubmed (ACC)

Dataset: PPI (micro-F1)