unit8co / darts

A python library for user-friendly forecasting and anomaly detection on time series.
https://unit8co.github.io/darts/
Apache License 2.0
8.04k stars 874 forks source link

Add iTransformer (Inverted Transformer) to DNN models #2138

Open ivelin opened 10 months ago

ivelin commented 10 months ago

Is your feature request related to a current problem? Please describe. It appears that DNN models are advancing on a monthly basis. The iTransofmer paper from Dec 2023 claims to be a more resource efficient model with better accuracy than TiDE, PatchTST, TimesNet and other recent DNNs for time series.

Describe proposed solution Consider adding iTransformer to the darts DNN model set.

Additional context iTransformer paper: https://arxiv.org/pdf/2310.06625.pdf

iTransformer implementation: https://github.com/thuml/iTransformer

lamkhatinh commented 9 months ago

+1

strakehyr commented 9 months ago

I wonder what performance we'd get to pairing this up with a symmetric layer of reversible instance normalization...