End‑to‑end GNN traffic forecasting on PeMS sensor data
The project provides a complete workflow for spatiotemporal traffic prediction using GraphSAGE on PeMS data, handling cleaning, imputation, graph construction, and weather alignment. It leverages PyTorch Geometric’s NeighborSampler and Distributed Data Parallel to scale training across multiple GPUs with mixed‑precision. Researchers and engineers can reproduce experiments via the Evidence Pack reporting system. Compared to simple baselines, it offers a ready‑to‑run, high‑performance solution for large‑scale traffic forecasting.
View on GitHub →Gonglz/pems-gnn-traffic-forecasting