HyperGALE

HyperGALE is the open source implementation of IJCNN accepted paper HyperGALE: ASD Classifcation via Hypergraph Gated Attention with Learnable Hyperedges

Dataset

Download the ABIDE-2 dataset from here.

Resting state fMRI data is preprocessed using Version 1 of Schaefer2018 parecellation with #ROIs=400. Details about about parcellations can be found here.

Usage

1. Change the path attribute in file source/conf/dataset/fc_abide2.yaml to the path of your dataset.

2. Run the following command to train the model.

python -m source --multirun model=hypergale,hypergraphgcnv2,hypergraphgcn,gcn,gat,graphsage dataset=fc_abide2 repeat_time=5

• model, default=(hypergale,hypergraphgcnv2,hypergraphgcn,gcn, gat, graphsage). Which model to use. The value is a list of model names.

• repeat_time, default=5. How many times to repeat the experiment. The value is an integer. For example, 5 means repeat 5 times.

Installation

conda create --name hypergel python=3.11
pip install torch
pip install pytorch-lightning
pip install torch-sparse torch-cluster torch-geometric

pip install pandas
pip install scikit-learn
pip install sicipy
pip install sympy
pip install matplotlib
pip install seaborn

pip install nilearn
pip install hydra-core
pip install omegaconf
pip install wandb
pip install ipdb

Dependencies

• python=3.11
• cudatoolkit=11.10
• torch==2.0.1
• torch-cluster==1.6.1
• torch-geometric==2.3.1
• torch-sparse==0.6.17
• torchmetrics==1.1.0
• torchvision==0.15.2
• pytorch-lightning==2.0.7
• numpy==1.24.2
• pandas==2.0.1
• scikit-learn==1.2.2
• scipy==1.10.1
• seaborn==0.12.2
• sympy==1.11.1
• nilearn==0.10.1
• hydra-core==1.3.2
• omegaconf==2.3.0
• wandb==0.15.0
• ipdb==0.13.13