fagcn

Frequency Adaptation Graph Convolutional Networks (FAGCN)

• Paper link: https://arxiv.org/abs/2101.00797
• Author's code repo: https://github.com/bdy9527/FAGCN

Dataset Statics

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

Refer to Planetoid.

Results

TL_BACKEND="tensorflow" python fagcn_trainer.py --dataset cora --lr 0.01 --l2_coef 0.0005 --drop_rate 0.4 --hidden_dim 16 --eps 0.3 --num_layers 5
TL_BACKEND="tensorflow" python fagcn_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.0005 --drop_rate 0.4 --hidden_dim 16 --eps 0.4 --num_layers 4
TL_BACKEND="tensorflow" python fagcn_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.001 --drop_rate 0.4 --hidden_dim 16 --eps 0.1 --num_layers 6
TL_BACKEND="paddle" python fagcn_trainer.py --dataset cora --lr 0.01 --l2_coef 0.001 --drop_rate 0.6 --hidden_dim 16 --eps 0.2 --num_layers 3
TL_BACKEND="paddle" python fagcn_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.001 --drop_rate 0.6 --hidden_dim 16 --eps 0.2 --num_layers 5
TL_BACKEND="paddle" python fagcn_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.001 --drop_rate 0.4 --hidden_dim 16 --eps 0.2 --num_layers 6
TL_BACKEND="torch" python fagcn_trainer.py --dataset cora --lr 0.005 --l2_coef 0.0005 --drop_rate 0.4 --hidden_dim 16 --eps 0.3 --num_layers 5
TL_BACKEND="torch" python fagcn_trainer.py --dataset citeseer --lr 0.005 --l2_coef 0.001 --drop_rate 0.4 --hidden_dim 16 --eps 0.3 --num_layers 3
TL_BACKEND="torch" python fagcn_trainer.py --dataset pubmed --lr 0.005 --l2_coef 0.001 --drop_rate 0.4 --hidden_dim 16 --eps 0.5 --num_layers 6

Dataset	Paper	Our(tf)	Our(pd)	Our(torch)
cora	84.1±0.5	83.1±0.4	82.1±0.4	78.1±0.7
citeseer	72.7±0.8	68.3±0.8	68.2±0.8	65.3±1.3
pubmed	79.4±0.3	79.2±0.1	79.7±0.3	77.9±0.8