make future dataframe ignores periods keyword

[thanosvlo]

Hi just a quick question on what is the reasoning behind this line :

neural_prophet/neuralprophet/forecaster.py

Line 2808 in 1b44634

periods = self.n_forecasts

Id assume - based on the documentation- that periods control the number of n_forecasts we want to predict into the future.
Say if the model is trained to predict 7 days into the future and the periods is set to be 3 then we would want to have a prediction for the upcoming 21 days in a rolling inference fashion.

are there any plans to amend the function to be in line with the documentation and offer the described functionality ?

Thank you for maintaining this repo, much appreciated.

[judussoari]

Hi @thanosvlo
When we use autoregression (i.e. self.max_lags > 0) and call m.make_future_dataframe() to predict into the unknown future, then we want to predict a defined number of steps. The number, in that case, is determined by n_forecasts.
We add the parameter period for the case of not using autoregression self.max_lags = 0. In this case, we allow the user to forecast periods steps into the unknown future.
periods and n_forecasts are not multiplied. Hope this helps!