neuralforecast icon indicating copy to clipboard operation
neuralforecast copied to clipboard

[Transformers] Improve exogenous covariates capabilities

Open cchallu opened this issue 1 year ago • 1 comments

Description

Current transformer models (PatchTST, Informer, Autoformer, FEDformer) only allow for calendar covariates. Adding the capability to include continuous temporal or static covariates would improve their performance on many domains where these types of covariates are useful.

Use case

No response

cchallu avatar Jun 08 '23 22:06 cchallu

Hi Cristian@cchallu, thanks for bringing this issue up. Does your library allow my own datasets without a calendar or any covariates? May I know which methods would allow me to run with my own data without covariates?

lzhang18 avatar Sep 05 '23 05:09 lzhang18