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

Luis Huergo

10/04/2023, 5:35 PM
In the meantime I have found out that the error has to do with a data frame called 'fcsts_df' which gets joined with `futr_df`by "unique_id" and "ds" thus throwing an error if the number of rows of futr_df is less than the rows of fcsts_df. I still don't understand what is happening though and would be very thankful for any help.
c

Cristian (Nixtla)

10/04/2023, 5:59 PM
Hi @Luis Huergo! We added extra protections for the predict method with
futr_df
input. There might be an issue in the tutorial, we will fix it!
l

Luis Huergo

10/04/2023, 6:34 PM
Hi @Cristian (Nixtla) thank you for your reply. Do you have a source for the right way to make a prediction on the 1 step (even nowcast) horizon case?
j

José Morales

10/04/2023, 8:10 PM
I took a quick look at that example and the
ds
are integers, and the frequency is set to monthly, so there's probably no match in those dataframes (fcst and futr). The check verifies that if you trained until 2001-03-31 and you have month end as a frequency your futr_df starts at 2001-04-30 for example. The data in that example doesn't work at the moment but you should be able to predict one step ahead with data that follows that chronological order
l

Luis Huergo

10/04/2023, 8:21 PM
Thank you @José Morales, I will give it a try!
🙏 1