node2vec

node2vec: Scalable Feature Learning for Networks

• Paper link: https://arxiv.org/abs/1607.00653

• Author's code repo (in Tensorflow):

  https://github.com/aditya-grover/node2vec.

• Popular pytorch implementation: https://github.com/pyg-team/pytorch_geometric/blob/master/examples/node2vec.py.

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="tensorflow" python node2vec_trainer.py --lr 0.05 --embedding_dim 128 --p 4.0 --q 0.25 --walk_length 20 --window_size 10 --num_walks 10 --n_epoch 100 --dataset cora
TL_BACKEND="tensorflow" python node2vec_trainer.py --lr 0.05 --embedding_dim 128 --p 4.0 --q 0.25 --walk_length 20 --window_size 10 --num_walks 10 --n_epoch 100 --dataset citeseer
TL_BACKEND="tensorflow" python node2vec_trainer.py --lr 0.05 --embedding_dim 64 --p 2.0 --q 0.5 --walk_length 10 --window_size 5 --num_walks 10 --n_epoch 100 --dataset pubmed
TL_BACKEND="torch" python node2vec_trainer.py --lr 0.05 --embedding_dim 128 --p 4.0 --q 0.25 --walk_length 20 --window_size 10 --num_walks 10 --n_epoch 100 --dataset cora
TL_BACKEND="torch" python node2vec_trainer.py --lr 0.05 --embedding_dim 128 --p 4.0 --q 0.25 --walk_length 20 --window_size 10 --num_walks 10 --n_epoch 100 --dataset citeseer
TL_BACKEND="torch" python node2vec_trainer.py --lr 0.05 --embedding_dim 64 --p 4.0 --q 0.25 --walk_length 10 --window_size 5 --num_walks 10 --n_epoch 100 --dataset pubmed
TL_BACKEND="paddle" python node2vec_trainer.py --lr 0.05 --embedding_dim 128 --p 4.0 --q 0.25 --walk_length 20 --window_size 10 --num_walks 10 --n_epoch 100 --dataset cora
TL_BACKEND="paddle" python node2vec_trainer.py --lr 0.05 --embedding_dim 128 --p 4.0 --q 0.25 --walk_length 20 --window_size 10 --num_walks 10 --n_epoch 100 --dataset citeseer
TL_BACKEND="paddle" python node2vec_trainer.py --lr 0.05 --embedding_dim 64 --p 2.0 --q 0.5 --walk_length 10 --window_size 5 --num_walks 10 --n_epoch 100 --dataset pubmed

Dataset	pyg	Our(pd)	Our(torch)	Our(tf)
cora	71.06(±0.62)	72.33(±1.65)	71.91(±0.74)	71.61(±0.96)
citeseer	48.20(±1.01)	48.55(±1.32)	48.76(±0.91)	48.23(±1.35)
pubmed	61.60(±1.83)	62.28(±1.33)	61.82(±2.48)	62.89(±2.45)

Note: The hyperparameter settings for the results in pyg are the same as in GammaGL.