GammaGL/examples/chebnet

• Chebyshev Networks (ChebNet)
• Dataset Statics
• • Results

Chebyshev Networks (ChebNet)

• Paper link: https://arxiv.org/abs/1606.09375

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

# k=2, available dataset: "cora", "citeseer", "pubmed"
TL_BACKEND="paddle" python chebnet_trainer.py --k 2 --dataset cora --lr 0.01 --hidden_dim 64 --drop_rate 0.8 --l2_coef 0.0005
TL_BACKEND="tensorflow" python chebnet_trainer.py --k 2 --dataset cora --lr 0.01 --hidden_dim 64 --drop_rate 0.7 --l2_coef 0.0005
TL_BACKEND="torch" python chebnet_trainer.py --k 2 --dataset cora --lr 0.01 --hidden_dim 64 --drop_rate 0.7 --l2_coef 0.0005
TL_BACKEND="paddle" python chebnet_trainer.py --k 2 --dataset citeseer --lr 0.01 --hidden_dim 32 --drop_rate 0.9 --l2_coef 0.0005
TL_BACKEND="tensorflow" python chebnet_trainer.py --k 2 --dataset citeseer --lr 0.01 --hidden_dim 64 --drop_rate 0.9 --l2_coef 0.0005
TL_BACKEND="torch" python chebnet_trainer.py --k 2 --dataset citeseer --lr 0.01 --hidden_dim 64 --drop_rate 0.9 --l2_coef 0.0005
TL_BACKEND="paddle" python chebnet_trainer.py --k 2 --dataset pubmed --lr 0.01 --hidden_dim 64 --drop_rate 0.6 --l2_coef 0.0005
TL_BACKEND="tensorflow" python chebnet_trainer.py --k 2 --dataset pubmed --lr 0.01 --hidden_dim 64 --drop_rate 0.6 --l2_coef 0.0005

# k=3, available dataset: "cora", "citeseer"
TL_BACKEND="paddle" python chebnet_trainer.py --k 3 --dataset cora --lr 0.01 --hidden_dim 64 --drop_rate 0.8 --l2_coef 0.0005
TL_BACKEND="tensorflow" python chebnet_trainer.py --k 3 --dataset cora --lr 0.01 --hidden_dim 64 --drop_rate 0.7 --l2_coef 0.0005
TL_BACKEND="torch" python chebnet_trainer.py --k 3 --dataset cora --lr 0.01 --hidden_dim 64 --drop_rate 0.7 --l2_coef 0.0005
TL_BACKEND="paddle" python chebnet_trainer.py --k 3 --dataset citeseer --lr 0.01 --hidden_dim 64 --drop_rate 0.9 --l2_coef 0.0005
TL_BACKEND="torch" python chebnet_trainer.py --k 3 --dataset citeseer --lr 0.01 --hidden_dim 64 --drop_rate 0.9 --l2_coef 0.0005

K=2

Dataset	Paper	Our(pd)	Our(tf)	Our(torch)
cora	81.2	80.78±1.12	80.62±0.80	80.42±1.18
citeseer	69.6	69.45±1.75	70.34±1.46	70.58±0.62
pubmed	73.8	76.90±1.40	75.68±1.42	OOM

K=3

Dataset	Paper	Our(pd)	Our(tf)	Our(torch)
cora	79.5	81.88±1.32	80.55±1.75	81.72±1.38
citeseer	69.8	70.27±0.93	OOM	70.18±0.82
pubmed	74.4	OOM	OOM	OOM