dgl/examples/pytorch/P-GNN

DGL Implementations of P-GNN

This DGL example implements the GNN model proposed in the paper Position-aware Graph Neural Networks. For the original implementation, see here.

Contributor: RecLusIve-F

Requirements

The codebase is implemented in Python 3.8. For version requirement of packages, see below.

dgl 0.7.2
numpy 1.21.2
torch 1.10.1
networkx 2.6.3
scikit-learn 1.0.2

Instructions for experiments

Link prediction

# Communities-T
python main.py --task link

# Communities
python main.py --task link --inductive

Link pair prediction

# Communities
python main.py --task link_pair --inductive

Performance

Link prediction (Grid-T and Communities-T refer to the transductive learning setting of Grid and Communities)

Dataset	Communities-T	Communities
ROC AUC ( P-GNN-E-2L in Table 1)	0.988 ± 0.003	0.985 ± 0.008
ROC AUC (DGL: P-GNN-E-2L)	0.984 ± 0.010	0.991 ± 0.004

Link pair prediction

Dataset	Communities
ROC AUC ( P-GNN-E-2L in Table 1)	1.0 ± 0.001
ROC AUC (DGL: P-GNN-E-2L)	1.0 ± 0.000