Scalable GNN Training via Parameter Freeze and Layer Detachment

Accepted by DASFAA 2025

How to Run

For dataset Flickr, Reddit, Yelp and Amazon, download them from Google Drive link or BaiduYun link(code:f1ao) and put them at correct place. Only four files are needed: adj_full.npz, class_map.json, feats.npy and role.json. These public datasets are collected by GraphSAINT and are irrelevant to this paper.

/PATH/TO/THIS_DIR
|   
└───mensa
| 
└───dataset/
|   |   flickr/
|   |   reddit/
|   |   ...

For Ogbn-arxiv and Ogbn-products, they will be downloaded automatically.

If you want to run algorithm mensa:

cd mensa
bash run.sh

Results will be saved at log directory.