Graph Nets
DeepMind's library for building graph networks in TensorFlow and Sonnet, enabling graph-structured data processing with neural networks.
What is Graph Nets?
Graph Nets is DeepMind's library for building graph networks in TensorFlow and Sonnet. It enables developers and researchers to create neural network models that operate directly on graph-structured data, where inputs and outputs are graphs with updated edge, node, and global attributes. This approach is particularly valuable for problems involving relational reasoning and structured data.
Machine learning researchers and developers working with graph-structured data, particularly those using TensorFlow for problems in relational reasoning, physical simulation, or combinatorial optimization.
Developers choose Graph Nets because it provides a clean, research-backed implementation of graph networks from DeepMind, with production-ready TensorFlow/Sonnet integration and comprehensive examples demonstrating practical applications across different domains.
Overview
Build Graph Nets in Tensorflow
Use Cases
Best For
- Implementing graph neural networks for relational reasoning tasks
- Processing graph-structured data in TensorFlow pipelines
- Building models for shortest path prediction in graphs
- Creating neural networks for physical system simulation
- Developing sorting algorithms using graph-based approaches
- Research on graph networks and relational inductive biases
Not Ideal For
- Projects using PyTorch, JAX, or other non-TensorFlow frameworks
- Teams requiring seamless integration with the latest TensorFlow versions without version constraints
- Applications needing a general-purpose graph neural network library with a large community and extensive pre-built models
Pros & Cons
Pros
Implements the graph network architecture from DeepMind's arXiv paper, providing a solid theoretical foundation for relational reasoning tasks.
Supports both TensorFlow 1 and 2 with corresponding Sonnet versions, allowing use in legacy and modern TF ecosystems, as detailed in the installation instructions.
Includes Jupyter notebooks for practical tasks like shortest path finding, sorting, and physical prediction, offering clear, hands-on examples to accelerate learning.
Provides tools for creating, transforming, and batching graph-structured data, simplifying data preparation for complex models.
Cons
Installation requires precise TensorFlow and Sonnet versions, with separate commands for CPU/GPU and TF1 vs TF2, increasing setup complexity and potential for errors.
Deeply integrated with TensorFlow and Sonnet, making it unsuitable for projects using other frameworks like PyTorch, and adding vendor dependency.
Centered on DeepMind's specific graph network model, which may not align with all GNN approaches or require extra work for customization beyond the provided modules.
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.
PyTorch-BigGraphGenerate embeddings from large-scale graph-structured data.
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.