search

facebook / prophet

Tool for producing high quality forecasts for time series data that has multiple seasonality with linear or non-linear growth.
https://facebook.github.io/prophet
MIT License
18.09k stars 4.49k forks source link

How extra regressors are forecasted in fbprophet in cross-validation? #1967

Open newUserName68 opened 3 years ago

newUserName68 commented 3 years ago

During validation of the model I have questions:

  1. Having initial = '1000days' why in df_cv the first forecast point starts from 1005 point?
  2. If I add extra regressor, how is it forecasted? I tried to use true values (which actually should not be used in cross-validation) and I tried to use the last true value from training set. But still I can't get the same yhat as in df_cv while making one separate forecast

P.S. I am not interested in 'for' loop, just wish to get an explanation Thanks in advance!

tcuongd commented 2 years ago

If I add extra regressor, how is it forecasted? I tried to use true values (which actually should not be used in cross-validation)

The current cross_validation function does actually use the true values for the extra regressors during validation, even though that is not the correct way to do it as you said.

Are you able to provide some sample data and code to reproduce this issue?