square-coder/STGODE

STGODE is a research-oriented implementation of a Spatial-Temporal Graph ODE network. While it was innovative at the time of its release (circa 2019-2020) by applying Neural ODEs to the spatial-temporal domain to capture continuous dynamics, it currently functions primarily as a static reference implementation for a specific academic paper. With 127 stars and a 'frozen' velocity (0 updates recently), it lacks the community momentum or software engineering rigor to serve as a production-grade library. In the competitive landscape of time-series and spatial-temporal forecasting, the field has rapidly moved toward Transformer-based architectures (e.g., Informer, Autoformer) and, more recently, large-scale foundation models (e.g., Lag-Llama, Chronos, or Google's TimesFM). The displacement horizon is set to 6 months because the state-of-the-art has already moved past this specific architectural niche for most practical applications. The moat is non-existent beyond the specific algorithmic contribution, and frontier labs are likely to supersede this via general-purpose temporal models rather than specialized GNN-ODE hybrids.