HGAMLP: A Scalable Training Framework for Heterogeneous Graph Neural Networks

OGBN-MAG result reproduction version

Requirements

Neural network libraries for GNNs

• Python 3.7.10
• dgl==0.8.2
• ogb==1.3.6
• torch==1.12.1+cu102
• torch_sparse==0.6.15

Data preparation

The pre-trained embeddings we use is the Line[1] method in RpHGNN[2]. To reproduce the results on the OGB Leaderboards (ogbn-mag), follow the steps below:

• For OGBN-MAG, the code will automatically download it via the ogb package.
• Download the pre-trained embeddings mag.p directly from Google Drive

References:

• [1] Jian Tang, Meng Qu, Mingzhe Wang, Ming Zhang, Jun Yan, and Qiaozhu Mei. "Line: Large-scale information network embedding." In Proceedings of the 24th international conference on world wide web, pp. 1067-1077. 2015.
• [2] RpHGNN https://github.com/CrawlScript/RpHGNN

For experiments on the large dataset ogbn-mag, the dataset will be automatically downloaded from OGB challenge.

Run HGAMLP for OGB Leaderboards (ogbn-mag)

python main.py --stages 400 400 400 500 --extra-embedding Line --num-hops 2 --label-feats --num-label-hops 2 --n-layers-1 2 --n-layers-2 2 --n-layers-3 3 --act leaky_relu --bns --label-bns --lr 0.002 --weight-decay 0 --threshold 0.75 --patience 100 --gama 10 --amp --seeds 0 1 2 3 4 5 6 7 8 9 --gpu 0