Biased Random-Key Genetic Algorithm for Optimizing Graph Neural Network Architectures
This repository implements a Biased Random-Key Genetic Algorithm (BRKGA) to optimize Graph Neural Network (GNN) architectures.
The approach searches over different GNN topologies by encoding architecture hyperparameters as chromosomes, which are decoded into models and evaluated on node classification tasks.
- BRKGA: Contains the BRKGA, CudaAwareBRKGA, and Decoder-GNN classes, which are used to run the experiments.
- DATASETS: Dataset support for Cora, CiteSeer, and PubMed.
- EXPERIMENTS: Ready-to-run experiment script (
run_cora.py).
- Andersson Alves da Silva (andersson.alves.silva@outlook.com)
- Ricardo Martins de Abreu e Silva (rmas@cin.ufpe.br)
Federal University of Pernambuco (UFPE), Recife, Pernambuco, Brazil
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.
pip install -r requirements.txt