LMissher/STGNN

STGNN is a five-year-old reference implementation of a specific spatial-temporal architecture. While it maintains a baseline level of interest (166 stars), it lacks the momentum or ecosystem of more modern frameworks. In the time since this repo was active, the field of spatio-temporal forecasting has shifted toward more sophisticated architectures like GMAN, MTGNN, and Spatio-Temporal Transformers. Defensibility is low because this is primarily a standalone algorithm implementation without a library structure, package distribution (no pip install), or ongoing maintenance (0.0 velocity). It faces high market consolidation risk from comprehensive time-series and graph libraries like 'Darts', 'Nixtla', or 'PyTorch Geometric', which offer more robust, maintained versions of similar concepts. Frontier labs are unlikely to compete directly as this is too domain-specific (traffic), but the underlying technology has already been largely superseded by newer SOTA models in the academic and industrial forecasting space.