Long-term Forecasting with TiDE: Time-series Dense Encoder - Unofficial Pytorch Implementation - WIP

Commands

Install

• make install

Data preparation

• make datasets

Metrics and wandb link with commands

name	test mae	paper mae	test mse	paper mse
ETTh1 96	0.450671	0.398	0.427044	0.375
ETTh1 192	0.486024	0.422	0.47277	0.412
ETTh1 336	0.527694	0.433	0.527147	0.435
ETTh1 720	0.60546	0.465	0.644379	0.454
ETTh2 96	0.284329	0.336	0.169457	0.27
ETTh2 192	0.318749	0.38	0.206157	0.332
ETTh2 336	0.337509	0.407	0.226434	0.36
ETTh2 720	0.400112	0.451	0.299051	0.419
ETTm1 96	0.369346	0.349	0.31828	0.306
ETTm1 192	0.399559	0.366	0.365496	0.335
ETTm1 336	0.429598	0.384	0.408837	0.364
ETTm1 720	0.470707	0.413	0.45856	0.413
ETTm2 96	0.225635	0.251	0.111662	0.161
ETTm2 192	0.248651	0.289	0.136481	0.215
ETTm2 336	0.271058	0.326	0.161007	0.267
electricity 96	0.236805	0.229	0.136566	0.132
electricity 192	0.251745	0.243	0.151502	0.147
electricity 336	0.283281	0.261	0.176307	0.161
electricity 720	0.307335	0.294	0.205603	0.196
traffic 96	0.280911	0.253	0.438414	0.336
traffic 192	0.284845	0.257	0.434229	0.346
traffic 336	0.290309	0.26	0.498862	0.355
traffic 720	0.309352	0.273	0.50618	0.386
weather 96	0.225352	0.222	0.161776	0.166
weather 192	0.256947	0.263	0.192686	0.209
weather 336	0.289164	0.301	0.224491	0.254
weather 720	0.330308	0.34	0.266972	0.313

ToDo

• ➕➖ Metrics reproducibility
  • some issues with traffic and etth1
• REVIN integration