FelixOpolka/STGCN-PyTorch

The project is a faithful PyTorch implementation of a seminal 2017 paper. While it has earned 370 stars, its velocity is effectively zero, and its age (7+ years) indicates it is a legacy research artifact rather than an evolving product. Its defensibility is low because the STGCN architecture is now considered a baseline in the spatio-temporal graph domain, having been surpassed by more advanced models like ASTGCN and Graph WaveNet. Modern graph learning frameworks like PyTorch Geometric (PyG) or DGL, and specialized libraries like LibCity, offer more robust, maintained, and optimized versions of these capabilities. Frontier labs are unlikely to build this specific niche tool, but they are increasingly releasing general-purpose time-series foundation models (e.g., Google's TimesFM, Amazon's Chronos) which may eventually render specialized graph architectures like this obsolete for many users. The primary risk is not from frontier labs, but from the natural evolution of open-source research where this project has already been 'placed on a shelf' as a reference implementation.