NFGNN

🚩 🚩 🚩 This is the Pytorch implementation of NFGNN (Node-oriented Spectral Filtering for Graph Neural Networks)!

Methodology

❤️NFGNN_large is corresponding to the scalable variant of NFGNN introduced in Sect.4.3.

❤️NFGNN is the standard version that includes the settings of transductive and inductive node classification.

Package dependencies

The project is built with Python3.6, CUDA11.2, NVIDIA GeForce RTX 3090. For package dependencies, you can install them by:

pip install -r requirements.txt

To start

Plz put the downloaded data or your own data in the folder below:

|-- root
    |-- NFGNN_standard
    |-- NFGNN_large
    |-- data
    |   |-- your own data

You can run the command to obtain the result of NFGNN:

python ./NFGNN/train_model.py --dataset $dataset_name

👉 (Optional) You can also run the code with preset hyperparameters (We will provide the preset hyperparameter as soon as possible!):

python ./NFGNN_standard/reproduce.sh

👉 (Optional) For your own dataset, you can run the command below to search for the best combination of hyperparameters:

python ./NFGNN_standard/meta/meta.py