torch-geometric-graph-neural-networks

PyTorch Geometric (PyG) for graph neural networks. Node classification, graph classification, link prediction with GCN, GAT, GraphSAGE, GIN layers. Message passing framework, mini-batch processing, heterogeneous graphs, neighbor sampling for large-scale learning, model explainability. Supports molecular property prediction (QM9, MoleculeNet), social networks, knowledge graphs, 3D point clouds. For non-graph deep learning use PyTorch directly; for traditional graph algorithms use NetworkX.