[Transformers] Improve exogenous covariates capabilities

[cchallu]

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

[lzhang18]

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?

[marcopeix]

Exogenous features support depends on the model's architecture. To my knowledge, current models are implemented mirroring their capabilities from their original papers.