Reproduce “Investigating the effects of the patch size in Time Series Foundation 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
Jad Yehya, Andrei-Tiberiu Pantea, Aymeric Dieuleveut, Matthieu Kowalski, Thomas Moreau
The authors compare patch sizes in encoder-only and decoder-only time-series forecasters trained under matched within-family conditions. Encoder models perform best with smaller patches, while decoder models perform best overall at P=64 and favor different sizes across forecast horizons. On 97 GIFT-Eval tasks, encoder P=8 wins 57 MASE and 64 CRPS tasks; findings apply to tested configurations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.