Davidham3/STSGCN

STSGCN was a high-impact research project from AAAI 2020 that introduced a method to capture spatial and temporal correlations simultaneously rather than through separate modules. With nearly 450 stars and 100+ forks, it remains a recognized benchmark in the Spatio-Temporal Graph Neural Network (STGNN) literature. However, as an open-source asset, its defensibility is low (3) because it is a static research artifact with zero current velocity. The field has moved toward Transformer-based architectures (e.g., Spatial-Temporal Transformer Networks) and more recently to large-scale pre-trained models for time series (e.g., UniST). Frontier labs are unlikely to compete directly in this niche 'traffic forecasting' domain, but the technology is effectively superseded by newer implementations in frameworks like PyTorch Geometric or DGL. Its primary value today is as a baseline for academic comparison rather than a production-ready library, especially given its dependency on MXNet, which has significantly declined in industry adoption compared to PyTorch and JAX.