PriPro

This is the official repository of paper Learning Prioritized Node-wise Message Propagation in Graph Neural Networks.

Setup

• torch==1.10.1
• torch-geometric==2.0.2
• networkx==2.5.1
• scipy==1.5.4
• numpy==1.19.2

Repository structure

├── datas  # old datasets, including cora
├── high_freq # new datasets, including texas, wisconsin, cornell, actor and chameleon
├── L2B
│   ├── exp.sh # run sh
│   ├── L2B_APPNP.py # model implementations
│   ├── loss.py # loss function
│   ├── run # Save run results
│   ├── train.py # the main code
│   └── utils.py # preprocessing subroutines
├── L2U
│   ├── exp.sh # run sh
│   ├── L2U_APPNP.py # model implementations
│   ├── loss.py # loss function
│   ├── run # Save run results
│   ├── train.py # the main code
│   └── utils.py # preprocessing subroutines

Run pipeline

1. You can run our model (L2B/L2U) by the script:

python PriPro/L2B/train.py --dataset cora

2. For more experiments running details, you can ref the running .sh by the script:

sh PriPro/L2B/exp.sh