This message was deleted.

Hey. The horizon is in "time steps" so if you provided data at the minute level

h=60

means 60 minutes, so if you actually want 30 * 24 * 60 steps ahead you should set that as the horizon. If you're only interested in the one hour you can just filter them out from the result, but they're needed for computing the predictions

Hi Jose, thanks for looking at these again! I'm shooting for the one hour horizon only, but I'd like to know the one hour forecast for every minute in the next month. Filtering out the 1-hour forecast from the results of

nf.predict()

I'd get 100 numbers, but I need 100 numbers for each timestamp in the one month test data (hence the 30 x 24 x 60 x 100 figure).

The 100 is the number of ids?

Yes, correct, I failed to explain I have 100 ids. Exactly, I'd need my 60 periods forecast, then move one period ahead and get the next 60 periods prediction, and so on.

I think you can do this with cross_validation by setting h=60 and step_size=1