gat

Graph Attention Networks (GAT)

• Paper link: https://arxiv.org/abs/1710.10903
• Author's code repo (in Tensorflow): https://github.com/PetarV-/GAT.
• Popular pytorch implementation: https://github.com/Diego999/pyGAT.

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="paddle" python gat_trainer.py --dataset cora --lr 0.01 --l2_coef 0.01 --drop_rate 0.7
TL_BACKEND="paddle" python gat_trainer.py --dataset citeseer --lr 0.006 --l2_coef 0.04 --drop_rate 0.6
TL_BACKEND="paddle" python gat_trainer.py --dataset pubmed --lr 0.05 --l2_coef 0.0015 --drop_rate 0.6
TL_BACKEND="torch" python gat_trainer.py --dataset cora --lr 0.01 --l2_coef 0.005 --drop_rate 0.7
TL_BACKEND="torch" python gat_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.01 --drop_rate 0.6
TL_BACKEND="torch" python gat_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.001 --drop_rate 0.2
TL_BACKEND="tensorflow" python gat_trainer.py --dataset cora --lr 0.01 --l2_coef 0.01 --drop_rate 0.7
TL_BACKEND="tensorflow" python gat_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.04 --drop_rate 0.7
TL_BACKEND="tensorflow" python gat_trainer.py --dataset pubmed --lr 0.005 --l2_coef 0.003 --drop_rate 0.6

Dataset	Paper	Our(pd)	Our(torch)	Our(tf)
cora	83.0(±0.7)	83.54(±0.75)	82.44(±0.43)	83.26(±0.96)
citeseer	72.5(±0.7)	72.74(±0.76)	70.94(±0.43)	72.5(±0.65)
pubmed	79.0(±0.3)	78.82(±0.71)	78.5(±0.75)	78.2(±0.38)