Would like to make use of past (observed variables) and future covariates (forecasts that are available before a forecast is made for that horizon). Examples:
- past: observed temperature: can use lagged data as inputs to the model (i.e. temperature on day N-1 to predict sales on day N)
- future: forecasts known ahead of prediction (i.e. could be that we have the weather forecast for day N)
| Date | forecast_available | observed_temperature | temperature_forecast | sales |
|---|---|---|---|---|
| 2021-01-01 | 2020-12-31 | 35 | 37 | 100 |
| 2021-01-02 | 2020-01-01 | 28 | 31 | 120 |
sktime abstracts the creation of lagged variables, etc, so any way to make the distinction between the two types of covariates?
