readme.md

Graph Attention Networks v2 (GATv2)

• Paper link: How Attentive are Graph Attention Networks?
• Author's code repo: https://github.com/tech-srl/how_attentive_are_gats.
• Annotated implemetnation: [https://nn.labml.ai/graphs/gatv2/index.html]

Dataset Statics

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

Results

TL_BACKEND="paddle" python gatv2_trainer.py --dataset cora --lr 0.01 --l2_coef 0.01 --drop_rate 0.7
TL_BACKEND="paddle" python gatv2_trainer.py --dataset citeseer --lr 0.0005 --l2_coef 0.01 --drop_rate 0.3
TL_BACKEND="paddle" python gatv2_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.001 --drop_rate 0.4

Dataset	Our(pd)	Our(tf)
cora	82.45(±0.34)	81.78(±0.29)
pubmed	70.9(±1.28)	69.9(±0.23)
citeseer	78.46(±0.19)	77.49(±0.08)