DR-GST

This repo is for source code of WWW 2022 paper "Confidence May Cheat: Self-Training on Graph Neural Networks under Distribution Shift".

Paper Link: https://arxiv.org/abs/2201.11349

Environment

• python == 3.8.8
• pytorch == 1.8.1
• dgl -cuda11.1 == 0.6.1
• networkx == 2.5
• numpy == 1.20.2

Main Methods

python main.py --dataset dataset --m model --labelrate labelrate --s stage --t threshold --b beta --drop_method drop_method --droprate droprate

• dataset: including [Cora, Citeseer, Pubmed, CoraFull, Flickr]
• model: including [GCN, GAT, APPNP]
• labelrate: including [3, 5, 10, 20]
• stage: ranging from 1 to 10
• drop_method: including [dropout, dropedge] e.g.

python main.py --dataset Citeseer --m GCN --labelrate 20 --s 9 --t 0.7 --b 0.5 --drop_method dropout --droprate 0.3

Part of Parameters

labelrate = 20, drop_method = dropout

Dataset-labelrate-threshold-beta-droprate
Cora-20-0.530000-0.333333-0.500000-dropout	stage=2
Citeseer-20-0.700000-0.500000-0.300000-dropout	stage=9
Pubmed-20-0.740000-1.000000-0.500000-dropout	stage=2
CoraFull-20-0.800000-0.250000-0.400000-dropout	stage=5
Flickr-20-0.890000-1.000000-0.100000-dropout	stage=9

labelrate = 20, drop_method = dropedge

Dataset-labelrate-threshold-beta-droprate
Cora-20-0.540000-0.500000-0.400000-dropedge	stage = 5
Citeseer-20-0.600000-0.500000-0.100000-dropedge	stage = 3
Pubmed-20-0.710000-1.000000-0.500000-dropedge	stage = 2
CoraFull-20-0.910000-0.200000-0.500000-dropedge	stage = 9
Flickr-20-0.950000-0.500000-0.200000-dropedge	stage = 6

labelrate = 10, drop_method = dropout

Dataset-labelrate-threshold-beta-droprate
Cora-10-0.700000-0.333333-0.300000-dropout	stage=6
Citeseer-10-0.550000-0.500000-0.300000-dropout	stage=4
Pubmed-10-0.750000-1.000000-0.300000-dropout	stage=1
CoraFull-10-0.780000-0.333333-0.100000-dropout	stage=6
Flickr-10-0.960000-1.000000-0.100000-dropout	stage=7

labelrate = 10, drop_method = dropedge

Dataset-labelrate-threshold-beta-droprate
Cora-10-0.650000-0.500000-0.500000-dropedge	stage = 3
Citeseer-10-0.650000-0.500000-0.300000-dropedge	stage = 6
Pubmed-10-0.600000-0.666667-0.300000-dropedge	stage = 1
CoraFull-10-0.630000-0.333333-0.300000-dropedge	stage = 2
Flickr-10-0.930000-0.666667-0.100000-dropedge	stage = 2

labelrate = 5, drop_method = dropout

Dataset-labelrate-threshold-beta-droprate
Cora-5-0.600000-0.500000-0.500000-dropout	stage=2
Citeseer-5-0.450000-0.666667-0.500000-dropout	stage=9
Pubmed-5-0.870000-1.000000-0.300000-dropout	stage=1
CoraFull-5-0.690000-0.333333-0.100000-dropout	stage=2
Flickr-5-0.930000-1.000000-0.100000-dropout	stage=2

labelrate = 5, drop_method = dropedge

Dataset-labelrate-threshold-beta-droprate
Cora-5-0.600000-0.500000-0.300000-dropedge	stage = 2
Citeseer-5-0.400000-0.500000-0.100000-dropedge	stage = 3
Pubmed-5-0.600000-1.000000-0.300000-dropedge	stage = 1
CoraFull-5-0.600000-0.500000-0.300000-dropedge	stage = 2
Flickr-5-0.960000-1.000000-0.500000-dropedge	stage = 4

labelrate = 3, drop_method = dropout

Dataset-labelrate-threshold-beta-droprate
Cora-3-0.700000-0.500000-0.300000-dropout	stage=3
Citeseer-3-0.450000-0.666667-0.300000-dropout	stage=2
Pubmed-3-0.780000-0.500000-0.500000-dropout	stage=1
CoraFull-3-0.870000-0.333333-0.300000-dropout	stage=2
Flickr-3-0.960000-1.000000-0.500000-dropout	stage=4

labelrate = 3, drop_method = dropedge

Dataset-labelrate-threshold-beta-droprate
Cora-3-0.400000-0.333333-0.100000-dropedge	stage = 2
Citeseer-3-0.450000-0.333333-0.500000-dropedge	stage = 1
Pubmed-3-0.810000-0.500000-0.300000-dropedge	stage = 1
CoraFull-3-0.660000-0.500000-0.300000-dropedge	stage = 1
Flickr-3-0.900000-0.666667-0.500000-dropedge	stage = 1

Contact

If you have any questions, please feel free to contact me with [email protected]