PyTorch Geometric (PyG) is an open-source library built on PyTorch for writing and training graph neural networks, which learn from data made of connected points.
In plain words
Some data is a network rather than a grid or a sentence: people who follow each other, atoms bonded in a molecule, roads between towns. A graph stores the things as nodes and the links as edges. PyG gives you a standard way to store graphs, ready-made graph neural network layers from published research, mini-batch loaders for many small graphs or one huge graph, and common benchmark datasets. It also handles 3D meshes and point clouds.
A quick example
The PyG introduction builds a tiny undirected graph of three nodes joined by four edges, where each node carries one number. You store the edges as a pair of lists saying which node links to which, wrap everything in a Data object, and can then pass it through a graph neural network layer.
Where you'll meet it
- Training on many small graphs or one giant graph.
- Graph learning research with common benchmark datasets.
- 3D meshes and point clouds.
Related
Sources
- DocsPyG Documentation, PyG · read 29 Sept 2026
- DocsIntroduction by Example, PyG documentation · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.