appnp

Predict then Propagate: Graph Neural Networks meet Personalized PageRank (APPNP)

• Paper link: Predict then Propagate: Graph Neural Networks meet Personalized PageRank

• Author's code repo:https://github.com/gasteigerjo/ppnp).

This example does not contain the implementation of PPNP.

Dataset Statics

Dataset	# Nodes	# Edges	# Classes
Cora	2,708	10,556	7
Citeseer	3,327	9,228	6
Pubmed	19,717	88,651	3
cora-ml	2810	7981	7

Results

TL_BACKEND="paddle" python appnp_trainer.py --dataset cora --lr 0.15 --n_epoch 200 --hidden_dim 64 --drop_rate 0.4 --l2_coef 0.02 --iter_K 6 --self_loops 0
TL_BACKEND="paddle" python appnp_trainer.py --dataset pubmed --lr 0.2 --n_epoch 250 --hidden_dim 64 --drop_rate 0.6 --l2_coef 0.001 --iter_K 10 --self_loops 2
TL_BACKEND="paddle" python appnp_trainer.py --dataset citeseer --lr 0.03 --n_epoch 500 --hidden_dim 32 --alpha 0.1 --drop_rate 0.4 --l2_coef 2e-3 --iter_K 10 --self_loops 1

dataset	paper	our(pd)	our(tf)
cora		82.9(±0.56)	76.3(±0.45)
citeseer	75.83(±0.27)	70.8(±0.00)	65.7(±0.15)
pubmed	79.73(±0.31)	79.8(±0.67)	75.5(±0)
cora-ml	85.29(±0.25)