We are trying to make predictions on time step > 1000
but when we calculate the diff_series we’re using future data (series[365:] will include this “unseen” data).
How can we use this to make real time predictions if we’re using future values?
Hi @EvilerEarth,
Thank you for your question. I hope I can help, though I am not sure I completely understand your doubt.
The idea of Differencing is that you subtract your past data (done by the -series[:-365] in the first line) to remove the spikes due to seasonality, then apply a moving average and re-introduce the seasonality adding again the “past” data.
This process should be done even for the “predicted” which are smoothed and then you re add the seasonality from the previous portion of data.
I hope I might have helped.
Best,
Davide