A modular and extensible framework for Graph Machine Learning (Graph ML) that supports multiple tasks at the node, edge, and graph levels.
The codebase integrates various Graph Neural Network (GNN) models, enabling researchers and developers to easily experiment, compare, and extend graph-based ML methods.
-
Multi-task Support
- Node-level: Classification, Regression, Clustering, Embedding
- Edge-level: Classification, Regression, Link Prediction
- Graph-level: Classification, Regression, Matching, Generation, Reconstruction
-
Integrated Models
- Graph Convolutional Network (GCN)
- Graph Attention Network (GAT)
- GraphSAGE
- Siamese GCN
- Graph Matching Networks (GMN)
- Graph Autoencoder (GAE)
- Graph Residual Autoencoder (GRAE)
- Graph Variational Autoencoder (GraphVAE)
- Graph Recurrent Neural Network (GraphRNN)
-
Dataset Loader & Preprocessing
- Built-in support for datasets like Cora, MUTAG, Karate Club, MovieLens 100K, and synthetic datasets.
- Feature normalization, adjacency matrix construction, and train-test splits.
-
Evaluation Metrics
- Classification: Accuracy, F1-score, Precision, Recall
- Regression: RMSE, MAE
- Graph Generation: Graph similarity scores, likelihood metrics
- Build a unified Graph ML framework for diverse graph-based tasks.
- Support easy integration of new models, datasets, and evaluation metrics.
- Provide reproducible experiments with configurable parameters.
- Facilitate both academic research and real-world applications.
This framework can be applied in multiple domains:
- Social Networks – Community detection, friend recommendations, fake account detection
- Healthcare – Disease diagnosis assistance, drug discovery, patient network analysis
- Transportation – Traffic prediction, route optimization, network resilience
- Fraud Detection – Suspicious transaction pattern detection
Benefits:
- Saves development time by providing a ready-to-use structure.
- Encourages ethical and transparent AI development.
- Reduces computational overhead by reusing modular components.