CF-HyperGNNExplainer

Project structure

• src_sparse/ — sparse variant (main entry points: src_sparse/train.py, src_sparse/main_explain.py).
• src_dense/ — dense variant (same entry points).
• models/ — Saved model checkpoints created by the training scripts.
• results/ — Generated counterfactual example pickles and evaluation outputs.

Installation and Setup

1. Prerequisites

• uv: This project uses uv for fast dependency management. Install it following the instructions at https://docs.astral.sh/uv/.
• Python >=3.14.

2. Install dependencies

Use uv to install the project's dependencies:

uv sync

3. Train the HGCN

cd src_sparse/
uv run train.py --dataset Cora --epochs 500

4. Generate the Counterfactual Explanations

uv run main_explain.py --dataset=cora --lr=0.1 --beta=0.5 --n-momentum=0.9 --cf-optimizer=SGD --strategy v1 --ckpt-path <path_to_pt_file>

5. Performance Evaluation

uv run evaluate.py --results <path_to_pkl_results_file> --strategy v1