Input Data (Shape) Error while running sf.forecast()

[ABJ66]

Hi all,

I am trying out various functionalities of statsforecast for univariate forecasting of multiple items. I'm running into this ValueError: ValueError: could not broadcast input array from shape (32,1) into shape (54,1).

Details for the setup/trial given as below:

1. Objective: forecast for multiple items which some of which have intermittent or have regular/non-intermittent history.
2. Data prepared in 3 column format: 'unique_id', 'ds', 'y'. Each item has exactly 54 time based measurements/values.
3. I've instantiated (a) StatsForecast class (as sf) with a bunch of models basis point # 1 and (b) asking for sf.forecast with h=6 and fitted = True, with input df basis point # 2.
4. It yields the ValueError: could not broadcast input array from shape (32,1) into shape (54,1)
5. Version 1.7.2 of statsforecast being used.
6. Earlier while using version 1.4.0, this error did not crop up, but the functionality of in-sample predictions was not implemented.

Would appreciate your inputs. Thank you in advance.

Regards.

[ABJ66]

Hi all,

FYI: I am using CrostonClassic as one of the methods basis point # 1 above. Seems to be a bug in this method when data points are having value of zero. I found it here: #772.

Pls let me know if this information is of use.

I would like to view/test the predictions on some portion of in-sample data as well. There is a Qn. on StackOverflow: https://stackoverflow.com/questions/77214782/how-to-get-fitted-valuesinsample-forecast-for-model-where-fitted-true-is-not-i

Thank you for making this framework it is really helpful to quickly test out various algorithms. Look forward to your comments/suggestions.

Regards.

[jmoralez]

Hey. Yes, it's because of that issue, we're going to make a small release tomorrow that will include a fix. Sorry for the troubles.