hcha

Hypergraph Convolution and Hypergraph Attention

• Paper link: https://arxiv.org/pdf/1901.08150.pdf

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

# available dataset: "cora", "citeseer", "pubmed"
# use attention by adding argument --use_attention=True
TL_BACKEND="paddle" python hcha_trainer.py --dataset cora --lr 0.01 --l2_coef 0.01 --drop_rate 0.9  
TL_BACKEND="paddle" python hcha_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.01 --drop_rate 0.7 
TL_BACKEND="paddle" python hcha_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.005 --drop_rate 0.6 
TL_BACKEND="torch" python hcha_trainer.py --dataset cora --lr 0.005 --l2_coef 0.01 --drop_rate 0.8 
TL_BACKEND="torch" python hcha_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.01 --drop_rate 0.7 
TL_BACKEND="torch" python hcha_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.002 --drop_rate 0.5 
TL_BACKEND="mindspore" python hcha_trainer.py --dataset cora --lr 0.01 --l2_coef 0.01 --drop_rate 0.6
TL_BACKEND="mindspore" python hcha_trainer.py --dataset citeseer --lr 0.01 --l2_coef 0.05 --drop_rate 0.7 
TL_BACKEND="mindspore" python hcha_trainer.py --dataset pubmed --lr 0.01 --l2_coef 0.01 --drop_rate 0.6 

Hypergraph Convolution

Dataset	Paper	Our(pd)	Our(th)	Our(ms)
cora	82.19	78.58±0.57	77.14±1.45	78.08±0.21
citeseer	70.35	63.32±1.55	63.48±1.39	62.98±0.71
pubmed	78.4	76.42±0.31	76.44±0.22	76.36±0.25

Hypergraph Attention

Dataset	Paper	Our(pd)	Our(th)	Our(ms)
cora	82.61	77.40±0.93	76.84±2.25	Err
citeseer	70.88	63.44±1.05	62.22±3.02	Err
pubmed	78.4	75.36±0.92	74.76±0.42	Err