Spectral Graph Neural Networks Meet Transformers (SpecFormer) Paper link: http://www.shichuan.org/doc/148.pdf Author's code repo: https://github.com/BUPT-GAMMA/Specformer. Note that the original code is implemented with Torch for the paper. Dataset Statics Dataset # Nodes # Edges # Classes Chameleon 2,277 36,101 5 Squirrel 5,201 217,073 5 Cora 2,708 10,556 7 Refer to WikipediaNetwork,Planetoid Results # available dataset: "chameleon", "squirrel", config below python spec_trainer.py --dataset=chameleon --n_epoch=500 --n_heads=4 --n_layer=2 --hidden_dim=32 --lr=0.001 --weight_decay=0.0005 --tran_dropout=0.2 --feat_dropout=0.4 --prop_dropout=0.5 python spec_trainer.py --dataset=squirrel --n_epoch=500 --n_heads=4 --n_layer=2 --hidden_dim=32 --lr=0.001 --weight_decay=0.001 --tran_dropout=0.1 --feat_dropout=0.4 --prop_dropout=0.4 python spec_trainer.py --dataset=cora --n_epoch=500 --n_heads=4 --n_layer=2 --hidden_dim=32 --lr=0.0002 --weight_decay=0.0001 --tran_dropout=0.2 --feat_dropout=0.6 --prop_dropout=0.2 Dataset Paper Our(th) chameleon 74.72±1.29 76.29±4.54 squirrel 64.64±0.81 65.31±1.55 cora 88.57±1.01 87.26±0.01