Hi, I am running a MlForecast model in which I have a long list of exogenous features, historical and dynamic to be precise. I do not have them at the time of forecasting. Is there a way to not pass them or do I need to use the lagged version at forecast time? Like imagine we are forecasting the amount of items sold, we know the price of an item now but we don’t know it in the future (I aggre you can but let’s assume you can’t). What can we do in such cases? Similar to Historical Exogenous Variables I see in NeuralForecast

Hey. You need to provide them at forecast time, we recently added a function in utilsforecast 0.2.4 that shifts the features by the horizon to make them "historic" (docs). Please let us know if this helps

Hello, I would need the exact inverse, other option is manually getting

horizon

steps from the past, get the exogenous and shift it to the future for again

horizon

steps. Do you think that makes sense?

like repeating the last horizon values?

Basically taking the last

horizon

entries from the training and shifting in time to cover the future

horizon

steps in the forecasting time

yeah that should work as well

@Hakan Ateşli FYI