Practical overview
When it fits
Use Graph Nets when a machine-learning task requires processing graph structure and updating edge, node or global attributes, or when exploring the project's graph-network demonstrations.
Who it is for
- Python developers building graph neural networks with TensorFlow and Sonnet.
- Researchers experimenting with learning tasks over relational or graph-structured data.
Documented uses
- Construct a graph-network module with custom edge, node and global update models.
- Train models on graph-structured tasks such as shortest-path finding, sorting and physical prediction.
- Study graph creation, manipulation and model construction through tutorial notebooks.
Requirements & limits
The repository is public and lists an Apache-2.0 license. The README describes local installation and notebook demos that can be run through Google Colaboratory.
- Graph data must be generated or provided by the developer; the example explicitly leaves this to the user.
- TensorFlow is not included as a declared library requirement and must be installed separately.
- The README describes both TensorFlow 1/Sonnet 1 and TensorFlow 2/Sonnet 2 dependency combinations.
- The included notebooks offer a browser-based Colaboratory route as well as a local Jupyter option.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Graph Nets with TensorFlow 2 and Sonnet 2
Install the library and its listed dependencies for the TensorFlow 2 and Sonnet 2 combination.
Before you start
- Python environment; the README lists Linux/macOS and Python 2.7 and 3.4+ for installation.
- Choose CPU or GPU TensorFlow packages.
Steps
For CPU, run: pip install graph_nets "tensorflow>=2.1.0-rc1" "dm-sonnet>=2.0.0b0" tensorflow_probability
For GPU, use the README's alternative command with tensorflow_gpu>=2.1.0-rc1 in place of tensorflow>=2.1.0-rc1.
Expected outcome
The documented package installation command installs Graph Nets alongside the selected TensorFlow package, Sonnet and TensorFlow Probability.
Guide limitations
- This is installation guidance; it does not cover training or running a model.
- The README's latest-version minimum is TensorFlow 1.15, and its TensorFlow 2 command specifies a 2.1.0 release candidate minimum.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- google-deepmind/graph_nets · Official homepage title · retained in its source language
Related options
jraph — A named graph neural network and graph-network library for JAX, listed in the repository README.
PyTorch Geometric — A named graph neural network library; the repository README points to its MetaLayer as an analogue.
Explore project →