Reproduce “A Critical Audit of Spatiotemporal Forecasting Benchmark Datasets and Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kenneth Martin, Simon Heilig, Asja Fischer, Michel F. C. Haddad, Adam M. Sykulski, Moshe Eliasof
The authors audit how data representations and temporal-only baselines affect spatiotemporal forecasting benchmark rankings. They compare graph-based models with autoregressive and residual-correction baselines, and examine preprocessing and heterogeneity in real and synthetic settings. On first-differenced Chickenpox, LSTM+ARIMA+Fourier records test MSE 0.713 versus 0.779 for GConvGRU. Results vary by task, so the audit does not show graph models are unnecessary.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.