Open jhug12 opened 5 months ago
Ciao @jhug12, thanks for the comment. Agree, it might just be a lags: int | None = None
. As outlined in the Discord, this implies working a bit more on functime's internals (of which I am not super familiar) so I can't work on it immediately. Any help is of course welcome.
Description:
I am proposing an enhancement for the
Forecaster
class in thefunctime
library, specifically to support settinglags=0
in classes that inherit fromForecaster
, such aslinear_model
. Currently, it is not possible to passlags=0
as an argument tolinear_model
or probably other similar models that derive from theForecaster
class.Issue:
The lack of support for
lags=0
restricts users from implementing a standard multiple regression approach within the forecasting framework provided byfunctime
. Here's an example of how one might attempt to set up a linear model forecaster without specifying lags, which implicitly requires non-zero lags:Desired Enhancement:
By allowing
lags=0
, users would be able to perform standard multiple regression analyses while benefiting from the features offunctime
, such as straightforward model changes and extensive cross-validation capabilities. This enhancement would make the library more versatile and applicable to a broader range of forecasting scenarios where traditional regression might be necessary.Use Case:
This feature would be particularly useful in scenarios where the target is not known for the last n days. One Case would be the the target is the difference from current temperature to future temperature (the change) over n days. If I want to look at the future change over two days then I wouldn't have a value for the last day. Sometimes directly predicting a future value is not possible as the time series is not stationary and therefore differencing is needed.