PyTorch Geometric Signed Directed
A PyTorch Geometric extension library for signed and directed graph neural networks, embedding, and clustering methods.
What is PyTorch Geometric Signed Directed?
PyTorch Geometric Signed Directed is an open-source extension library for PyTorch Geometric that provides specialized graph neural network models, data loaders, and utilities for analyzing signed and directed graphs. It solves the problem of applying deep learning to complex network structures where relationships have both direction (e.g., follower/following) and sign (e.g., trust/distrust), which are not handled by standard GNN libraries.
Researchers and practitioners in graph machine learning, network science, and data mining who need to work with signed social networks, directed citation graphs, trust networks, or any domain involving asymmetric and/or polarized relationships.
It consolidates numerous state-of-the-art research papers into a single, PyTorch Geometric-compatible package, saving users from re-implementing complex methods and providing a standardized interface for experimenting with signed and directed graph tasks.
Overview
PyTorch Geometric Signed Directed is a signed/directed graph neural network extension library for PyTorch Geometric. The paper is accepted by LoG 2023.
Use Cases
Best For
- Node classification in signed social networks (e.g., predicting user polarity)
- Link sign prediction in trust/distrust networks
- Clustering directed graphs based on flow imbalance
- Generating synthetic signed/directed graphs for benchmarking
- Applying spectral GNN methods using magnetic signed Laplacians
- Semi-supervised learning on signed networks with limited labels
Not Ideal For
- Projects focusing solely on undirected, unsigned graphs where standard GNN libraries suffice
- Production systems requiring optimized, out-of-the-box models with minimal tuning
- Teams lacking expertise in graph theory or recent GNN research papers
- Environments with strict dependency constraints or limited GPU memory
Pros & Cons
Pros
Implements over 20 state-of-the-art signed and directed GNN models from papers like SDGNN, MagNet, and DiGCL, saving users from re-implementing complex algorithms.
Built as an extension to PyTorch Geometric, it uses familiar data structures and workflows, ensuring compatibility with existing PyTorch codebases.
Includes data loaders for real-world datasets, specialized classes like SignedData, and preprocessing tools such as magnetic Laplacian computation for spectral methods.
Shows active CI testing, code coverage, and detailed documentation on ReadTheDocs, indicating reliable project health and support.
Cons
Models are often direct implementations from academic papers with minimal abstraction, requiring deep understanding of graph theory and extensive hyperparameter tuning.
Focused on experimentation rather than deployment, lacking features like batch processing optimizations or streamlined APIs for scalable inference.
Specialized for signed and directed graphs, making it overkill for general graph tasks and limiting community support compared to broader GNN libraries.
Frequently Asked Questions
Related Projects
GitHub repositoryGraph Neural Network Library for PyTorch
TRLTrain transformer language models with reinforcement learning.
dglPython package built to ease deep learning on graph, on top of existing DL frameworks.
Graph NetsBuild Graph Nets in Tensorflow
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.