The open source code for ICDM2022 paper "Unifying Graph Contrastive Learning with Flexible Contextual Scopes"

Overview

Our implementation for UGCL (Unfiying Graph Contrastive Learning) is based on PyTorch.

To run on cora/citeseer/pubmed

cd train_cora_citeseer_pubmed
python train_gp.py --data 'cora' --epoch 250 --patience 50 --lr 0.0002 --n 9 --hidden 4196 --sample_size 1500 #for cora
python train_gp.py --data 'citeseer' --epoch 150 --patience 50 --lr 5e-05 --n 2 --hidden 8192 --sample_size 3000 #for citeseer
python train_gp.py --data 'pubmed' --epoch 300 --patience 50 --lr 0.0002 --n 14 --sample_size 7000 --hidden 8192 #for pubmed

To run on ogbn-arxiv

cd train_ogbn
python train_ogbn --data 'ogbn' --n 4 --sample_size 2000 --hidden 8192 --test_epoch 2000 --patience 1000