Hi, I want to create a feature that counts the days of since last non-zero. Currently I'm creating the feature manually using pandas and I realized that I can't use it on MLForecast when forecasting. In production, We will not have test data, thus we don't the data of the latest days value. I think that I need to create a lag transform so that MLForecast can help put the values in

X_df

automatically when doing

predict

. My question is: • Is the statement correct that this can be handled by MLForecast on

predict

if it is defined in lag_transform? • Can njit in lag transform used to do something like this?

hey. yes, you can write your logic in a function decorated with njit and provide that through lag_transforms, that will take care of computing the feature and updating it during predict

All right, thank you for the answer!