https://github.com/nixtla logo
s

Stefan Wiegand

08/30/2023, 1:03 PM
Dear Nixtla-Team, in your Exogenous Variable Notebook (https://github.com/Nixtla/neuralforecast/blob/main/nbs/examples/Exogenous_Variables.ipynb) you use NHITS but from the NHITS Paper (https://arxiv.org/pdf/2201.12886.pdf) I understand that it does not make use of Exogenous Variables: "Our model, operating in the univariate regime and accepting only the predicted time-series’ history,". Have you just used the wrong model to demonstrate Exogenous Vars or am I misunderstanding the paper?
k

Kin Gtz. Olivares

08/30/2023, 1:51 PM
Hi @Stefan Wiegand For a fair comparison in the NHITS paper we only compare with univariate methods because most of the Transformer methods from the forecasting literature are restrict to univariate inputs (no exogenous).
c

Cristian (Nixtla)

08/30/2023, 3:25 PM
Complementing Kin's answer, while we didn't include exogenous in the paper, the implementation of the
NHITS
for general use in the library allows for exogenous variables
s

Stefan Wiegand

08/31/2023, 7:29 PM
Great, thanks for making this clear.
2 Views