A Coding Implementation on Spatial Graph Neural Networks for Urban Function Inference Using city2graph, OSMnx, and PyTorch Geometric

This tutorial outlines an end-to-end spatial graph learning pipeline using city2graph for urban function inference. The process begins with collecting real urban Point of Interest (POI) data and street network information from OpenStreetMap, with a synthetic data fallback for workflow reliability. This initial step ensures a robust dataset for analysis, regardless of online data availability.

Following data acquisition, the tutorial focuses on engineering spatial features and constructing multiple proximity graph families. It compares different graph-building strategies to represent the same urban environment effectively. This stage is crucial for understanding how spatial relationships influence urban patterns.

The pipeline then progresses to creating both heterogeneous and homogeneous graph structures, converting them into PyTorch Geometric format. A GraphSAGE model is subsequently trained to predict POI categories based on spatial structure. This demonstrates the application of graph neural networks (GNNs) in interpreting complex urban data.

The overall workflow integrates geospatial data processing, graph construction, and GNN-based urban function inference. This comprehensive approach provides a practical method for analyzing and understanding urban environments through their spatial and functional characteristics.