Generates h future rows from the task's skeleton (using generate_newdata()), optionally overlays user-supplied
newdata onto those rows, and predicts with the trained learner via mlr3::Learner$predict_newdata(). Works with
RecursiveForecaster, DirectForecaster, and classic LearnerFcst* forecasters.
Usage
# S3 method for class 'Learner'
forecast(object, task, h = 12L, newdata = NULL, ...)Arguments
- object
(mlr3::Learner)
A trained forecast learner.- task
(TaskFcst)
Provides the metadata needed to construct future rows: the order column (to extend the time index), key columns (for keyed tasks),freq, and the column-type schema expected bypredict_newdata(). The task's data values are not used. Pass the training task or any other schema-compatible TaskFcst.- h
(
integer(1))
Forecast horizon — number of future time steps per key.- newdata
(
data.frame()|NULL)
Optional exogenous features for future rows. Must contain the order column (and any key columns for keyed tasks), and every row must match a row of the generated future grid. Columns other than those are overlaid onto the generated skeleton, while skeleton rows without a match keepNA.- ...
(any)
Ignored.