NODE-SELECT Graph Neural Networks

This software package implements the work done in the paper "NODE-SELECT: A Graph Neural Network Based On A Selective Propagation Technique". This is the official Pytorch repository.

Installation

Install any of the relevant packages if not already installed:

• Pytorch (tested on 1.4.0) - preferably version 1.4.0 or later
• Numpy (tested on 1.19.2)
• Pandas (tested on 1.1.2)
• Scikit-learn (tested on 0.23.2)
• Matplotlib (tested on 3.2.2)
• PyTorch-Geometric (tested on 1.4.3)
• Tabulate (tested on 0.8.7)

• Pytorch, Numpy, Pandas, Scikit-learn, Matplotlib, and Tabulate

pip install torch torchvision 
pip install numpy
pip install pandas
pip install scikit-learn
pip install matplotlib
pip install tabulate

• PyTorch Geometric documentation

Usage

Once all the aforementionned requirements are satisfied, one can easily our codes using the appropriate flags. A set of the best hyper-parameters is provided in this file, please refer to it.

• Example-1. Train & evaluate a NODE-SELECT (NSGNN) model on the Pubmed dataset: with a learning-rate of 0.01, weight-decay of 0.0005, and built with 5 layers (filters).

python main.py --benchmark pubmed --framework NSGNN --lr 0.01 --weight_decay 0.0005  --layers 5

• Example-2. Train & evaluate a GCN model on the Cora dataset: with a learning-rate of 0.01, weight-decay of 0.0005, and built with 2 layers.

python main.py --benchmark cora --framework GCN --lr 0.01 --weight_decay 0.0005  --layers 2

• Example-3. Train & evaluate a GAT model on the Amazon-Photos dataset: with a learning-rate of 0.005, weight-decay of 0.00005, 128 neurons (16 neurons*8heads), built with 2 layers, and 8 attention-heads.

python main.py --benchmark amazon-p --framework GAT --lr 0.005 --weight_decay 0.00005  --layers 2 --heads 8 --neurons 16

Citation

If you find our codes or project useful for your research, please cite our work:

@article{louis2022node,
  title={Node-select: a graph neural network based on a selective propagation technique},
  author={Louis, Steph-Yves and Nasiri, Alireza and Rolland, Fatima J and Mitro, Cameron and Hu, Jianjun},
  journal={Neurocomputing},
  year={2022},
  publisher={Elsevier}
}