Hello
Is there a reference about Fourier transform prior to designing the model? Is it always good practice to find the hidden periods and tune the batch size around them? Week 4 mentions that a bit but with little details and in all the example the period is known beforehand. It is convenient but surely not always the case…
Fourier transform does not require you to know the period of seasonality in advance; it is used specifically to discover the dominant seasonal periods within your data.
I am not surely understanding your second part of completely. if you are asking before creating a model do we need to tune the seasonality in the time series data always then that would again depend on factors of frequency distribution between this period of seasonality, meaning if this hidden periods are in cycles and one is creating a model to understand this frequency i.r.t. to a third factor, then period of seasonality is must known factor to be addressed first in time series data.
But in case if this hidden period is specific to an event or a particular season, then knowing I advance wouldn’t be mandatory.