Rolling prediction ；validation model

[shandongpengyuyan]

My data was collected on a monthly basis from 2015 to 2022, which includes exogenous variables that vary over time. The model was trained using data from 2015 to 2021, with a prediction window of one month. The model was validated on the data from 2022 (12 step prediction). How to input the real predictive variables into the model and predict together with exogenous variables when making downward predictions.

[dennisbader]

Hi @shandongpengyuyan, I'm not 100% sure I understand your question.

This section in our Regression Model example notebook describes the difference between auto-regressive and one-shot forecasts, and also how exogenous variables are used (call past_covariates and future_covariates in Darts). The extraction of the correct covariates is done behind the hood by the models.