YuanchenBei / LRGA-vessel

This is the LRGA method for ogbl-vessel of the OGB Challenge.

LRGA-vessel

This is the LRGA solution for ogbl-vessel of the OGB Challenge.

Dependencies

python==3.8
torch==1.10.1+cu102
torch-geometric==2.0.4
ogb==1.3.4

OGB Dateset

The dataset ogbl-vessel should be download and placed in ./dataset/ogbl_vessel/.

Results on OGB-vessel

Performance on ogbl-vessel (10 runs):

Highest Train AUC: 50.64 ± 2.61

Highest Valid AUC: 54.18 ± 4.39

Final Train AUC: 47.84 ± 2.29

Final Test AUC: 54.15 ± 4.37

Run Command

python gnn.py --encoder_name='lrga' --hidden_channels=16 --lr=1e-6 --num_layers=3 --epochs=100 --dropout=0.5

Reference

[1] https://github.com/snap-stanford/ogb/tree/master/examples/linkproppred/vessel

[2] https://github.com/omri1348/LRGA