BlockGCL (Under review)

This is a PyTorch implementation of BlockGCL from the paper "Scaling Up, Scaling Deep: Blockwise Graph Contrastive Learning".

Requirements

• numpy==1.21.5
• torch==1.12.0
• torch-cluster==1.6.0
• torch_geometric==2.1.0.post1
• torch-scatter==2.0.9
• torch-sparse==0.6.15
• CUDA 11.6

Reproduction

To reproduce our results, please run:

bash run.sh

Due to the absence of predefined partitions for the Photo, Computer, CS, and Physics datasets, you should create a folder named "mask" in the current directory to store the random partitions.