pytorch_geometric_temporal

Why we included this project

Most time-series tooling ignores the graph underneath the data, and most graph libraries ignore time. PyTorch Geometric Temporal sits between the two: it extends PyTorch Geometric so you can train graph neural networks over sequences of graph snapshots, with implementations drawn from published papers covering traffic flow, epidemic spread, wind farm output, and similar spatio-temporal regression tasks. The unglamorous work is handled too, with snapshot iterators for static or dynamic graphs, train-test splitting, and bundled benchmark datasets like METR-LA and PEMS-BAY so you can check against published baselines. Researchers get a broad set of recent methods to build on, and practitioners get a working prototype faster than they would writing the batching and splitting code themselves. If you already work in PyTorch and think in graphs, this is a quick way to test a temporal-graph idea.