dcrnn

Simulation code for 'When Spatial Signal Is Not Enough: Learning Validity, Model Advantage, and Computational Feasibility in a Staged Simulation Study' (Saito & Ohira, 2026)

Reduced reproduction of spatio-temporal GNN traffic forecasting (GraphCast → DCRNN/Graph WaveNet) on METR-LA & PEMS-BAY — trained STGNN beats baselines, learned-adjacency ablation generalizes across two cities, on a consumer GPU