pytorch_geometric

Why we included this project

PyTorch Geometric is the library most teams reach for when they need to train neural networks on data that is naturally a graph, whether that means social networks, molecules, citation graphs, or 3D point clouds. It implements a wide range of graph neural network layers and models drawn from published papers, so you can start from a known architecture instead of writing message-passing code yourself. The practical side is covered too: mini-batch loaders that handle both many small graphs and a single giant one, multi-GPU training, and torch.compile support. It also bundles common benchmark datasets and transforms for graphs, meshes, and point clouds, which makes it a convenient base for research experiments and production pipelines. If you already work in PyTorch, the API will feel familiar, and the documentation and example notebooks get a first GNN running quickly.