Informer2020 无阉割分步复现:官方 Jupyter 版拆解为 .py 工程,ProbSparse Attention + 生成式解码,ETTh1 滑窗预测(seq96 / label48 / pred24),「6 轮早停 vs 30 轮不早停」对照实验定量验证早停价值(测试 R² 0.77 vs 0.45),踩坑与调参笔记全程留痕。
-
Updated
Jan 22, 2026 - Python
Informer2020 无阉割分步复现:官方 Jupyter 版拆解为 .py 工程,ProbSparse Attention + 生成式解码,ETTh1 滑窗预测(seq96 / label48 / pred24),「6 轮早停 vs 30 轮不早停」对照实验定量验证早停价值(测试 R² 0.77 vs 0.45),踩坑与调参笔记全程留痕。
Transformer-based multi-step time-series forecasting on ETTh1 - baselines, official LSTF 96->96 protocol, 5-seed statistics, tests and CI.
Do you need a transformer to forecast? Five tiny models (none over 80k params) vs published transformer results on the ETTh1 long-horizon benchmark — with an in-browser demo.
Launch apps, search files, and run commands instantly with a native macOS hotkey launcher that stays under 100 MB RAM.
GRU ile ETTh1 veri seti üzerinde transformatör yağ sıcaklığı tahmini
To associate your repository with the etth1 topic, visit your repo's landing page and select "manage topics."