lehaifeng/T-GCN

T-GCN is a highly influential academic project in the spatio-temporal forecasting niche, evidenced by its significant star count (1749) and high fork ratio (473). It established a standard pattern of combining GCNs with RNN variants to solve urban mobility problems. Its defensibility is rooted in its status as a 'gold standard' baseline for research; any new paper in this field must cite and compare against T-GCN. However, its zero velocity and age (2700+ days) indicate it is no longer being actively developed. Technically, it is being displaced by newer architectures such as Graph WaveNet, ASTGCN, and attention-based (Transformer) spatio-temporal models which handle long-range dependencies better than GRUs. While frontier labs like Google (via Google Maps) have far more sophisticated proprietary systems, they are unlikely to release specialized open-source tools that directly compete with this academic baseline. The primary risk is 'academic obsolescence'—it remains a benchmark, but is rarely the state-of-the-art choice for new production deployments.