mixhop

Higher-Order Graph Convolutional Architectures via Sparsified Neighborhood Mixing (MixHop)

• Paper link: https://arxiv.org/abs/1905.00067

• Author's code repo (in Tensorflow):

  https://github.com/samihaija/mixhop.

• Popular pytorch implementation: https://github.com/dmlc/dgl/tree/master/examples/pytorch/mixhop.

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 mixhop_trainer.py --dataset cora --lr 0.001 --l2_coef 1e-2 --drop_rate 0.5 --hidden_dim 64 --n_epoch 200
TL_BACKEND="paddle" python mixhop_trainer.py --dataset citeseer --lr 0.001 --l2_coef 5e-2 --drop_rate 0.4 --hidden_dim 64 --n_epoch 200
TL_BACKEND="paddle" python mixhop_trainer.py --dataset pubmed --lr 0.0005 --l2_coef 5e-4 --drop_rate 0.5 --hidden_dim 256 --n_epoch 200
TL_BACKEND="torch" python mixhop_trainer.py --dataset cora --lr 0.0005 --l2_coef 5e-4 --drop_rate 0.4 --hidden_dim 64 --n_epoch 200
TL_BACKEND="torch" python mixhop_trainer.py --dataset citeseer --lr 0.001 --l2_coef 5e-2 --drop_rate 0.4 --hidden_dim 64 --n_epoch 200
TL_BACKEND="torch" python mixhop_trainer.py --dataset pubmed --lr 0.001 --l2_coef 1e-3 --drop_rate 0.5 --hidden_dim 256 --n_epoch 200
TL_BACKEND="tensorflow" python mixhop_trainer.py --dataset cora --lr 0.005 --l2_coef 5e-3 --drop_rate 0.5 --hidden_dim 64 --n_epoch 200
TL_BACKEND="tensorflow" python mixhop_trainer.py --dataset citeseer --lr 0.001 --l2_coef 5e-2 --drop_rate 0.4 --hidden_dim 64 --n_epoch 200
TL_BACKEND="tensorflow" python mixhop_trainer.py --dataset pubmed --lr 0.001 --l2_coef 1e-3 --drop_rate 0.5 --hidden_dim 256 --n_epoch 200

Dataset	Paper	Our(pd)	Our(torch)	Our(tf)
cora	81.8(±0.62)	81.6(±0.57)	81.7(±1.1)	82.1(±0.90)
citeseer	71.4(±0.81)	71.4(±0.84)	71.3(±0.54)	71.5(±0.53)
pubmed	80.0(±1.1)	78.7(±0.46)	79.0(±0.39)	78.7(±0.27)