InterpretableGCN

Usage

Setup

The whole implementation is built upon PyTorch and PyTorch Geometric

conda

See the environment.yml for environment configuration.

conda env create -f environment.yml

PYG

To install pyg library, please refer to the document

Dataset

ADNI

We download this dataset from here. We treat multi-modal imaging scans as a brain graph.

How to run classification?

The BrainGNN framework is integrated in file main_braingnn.py. To run

python main_braingnn.py 

The Sparse Interpretable GNN framework is integrated in file main_sgcn.py. To run

python main_sgcn.py 

You can also specify the learning hyperparameters to run

python main_sgcn.py --epochs 200 --lr 0.0001 --search --cuda 0

main_sgcn.py: tunning hyperparameters

kernel/train_eval_sgcn.py: training framework for SGCN

kernel/train_eval_braingnn.py: training framework for BrainGNN

kernel/sgcn.py: training model for SGCN

kernel/braingnn.py: training framework for BrainGNN