RE-GNN

This repository contains a PyTorch implementation of our RE-GNN model in our paper "Scalable and Effective Graph Neural Networks via Trainable Random Walk Sampling".

Main Environments

• CPU: Intel(R) Xeon(R) Gold 8358 64C@2.6GHz.
• GPU: NVIDIA A100 40GB PCIe.
• CUDA toolkit: 11.3.
• Pytorch: 1.12.1.
• Pytorch-geometric: 2.1.0.
• OGB.

Datasets Preparation

Modify the variable datasetname and execute python preprocessor.py to initialize the datasets.

Optional datasetname are all listed in the file.

Required datasets will be downloaded automatically.

Reproduction

We provide some commands in script.sh to help you reproduce some results we report in Table 3, 4 of the paper.

Change --model and --dataset parameter for testing on other models and datasets.