GPCANet

Code for paper [Connecting Graph Convolutional Networks and Graph-Regularized PCA]

Supported datasets: cora, citeseer, pubmed, arxiv, products.
The last two are from OGB datasets. And products dataset needs --minibatch option.

run GPCA

Note that --freeze and --nlayer 1 are required for plain GPCA.

python main.py --data arxiv --model GPCANet --nlayer 1 --lr 0.1 --freeze --alpha 20 

run GPCANet

python main.py --data arxiv --model GPCANet --lr 0.005 --nlayer 3 --alpha 1 

run GCN (or GAT APPNP)

python main.py --data arxiv --model GCN --lr 0.005 --nlayer 3
python main.py --data arxiv --model GAT --lr 0.005 --nlayer 3
python main.py --data arxiv --model APPNP --lr 0.005 --nlayer 3

run GPCANet-initialized GCN

python main.py --data arxiv --model GCN --lr 0.005 --nlayer 3 --init

Bash scripts with hyperparameter search

See scripts folder.