https://github.com/nixtla logo
#neural-forecast
Title
# neural-forecast
a

Anna-Maria K.

07/18/2023, 3:59 PM
Hi all! I have a question. When I forecast I would like to forecast 30 days ahead. Thus, in last day of August, I want to forecast the whole month of October (thus 30 days). I want to use the cross_validation function to evaluate my results. I tried adjusting the n_windows, step_size and horizon, but I didn't get to the correct result. Any tips on how to implement time gap when using cross_validation function? Thanks a lot in advance!
c

Cristian (Nixtla)

07/18/2023, 4:21 PM
Hi @Anna-Maria K.! The library does not support his option currently, it will predict the immediate data after the train/validation set. The easiest solution is to increase the horizon to account for the gap. Be careful, however, that changing the
n_windows
and
step_size
are very different to the
horizon
. It seems that if you do not have September's data, you will need to have 1 window,
n_windows=1
, with a large horizon.
a

Anna-Maria K.

07/19/2023, 7:25 AM
Thanks a lot for the clarifications!