This project explores a simple yet effective approach for graph learning by combining multiple graph and node features into a single feature vector. Instead of relying only on Graph Neural Networks (GNNs), the project evaluates a Multi-Layer Perceptron (MLP) as a lightweight alternative across multiple benchmark datasets.
- Graph feature fusion through feature concatenation
- MLP-based graph classification
- Evaluation on multiple graph datasets
- Performance comparison across datasets
- Modular and easy-to-extend implementation
- Python
- PyTorch
- NumPy
- Scikit-learn
- Jupyter Notebook
- Load graph dataset.
- Extract multiple graph features.
- Concatenate features into a unified representation.
- Train an MLP classifier.
- Evaluate model performance.
Investigate whether a simple feature engineering approach combined with an MLP can achieve competitive performance on graph learning tasks while reducing model complexity.
Parisa Arbab
It is better to create a virtual environment for this project.
python3 -m venv <myenvpath>Install all required packages
pip install -r requirements.txt