Create a Direct Forecast Learner — direct

Function to create a DirectForecaster object. This is the recommended way to construct a direct forecaster. It is a thin wrapper around DirectForecaster$new().

A direct forecaster trains a separate regression model per forecast horizon, so predictions never feed back into one another (no error accumulation). For the recursive strategy (a single iterated model) see recursive_forecaster().

Usage

direct_forecaster(
  learner,
  lags,
  horizons,
  id = NULL,
  param_vals = list(),
  predict_type = NULL
)

Arguments

learner: (mlr3::Learner | mlr3pipelines::Graph | mlr3pipelines::PipeOp)
A regression learner or a graph/PipeOp (without PipeOpFcstLags).
lags: (integer())
The base lag values. Exposed in $param_set as lags, so it can be tuned via mlr3tuning::AutoTuner.
horizons: (integer())
Either a single integer H (expanded to 1:H) or an integer vector of specific horizons. One model is trained per horizon.
id: (character(1) | NULL)
Identifier, default NULL (auto-generated from the learner id).
param_vals: (named list())
Hyperparameter values applied to every horizon model. Per-horizon hyperparameters are not currently supported.
predict_type: (character(1) | NULL)
The predict type, default NULL.

Value

DirectForecaster.

Examples

library(mlr3pipelines)

task = tsk("airpassengers")
split = partition(task, ratio = 0.8)

# one model per horizon
flrn = direct_forecaster(lrn("regr.rpart"), lags = 1:3, horizons = length(split$test))
flrn$train(task, split$train)
flrn$predict(task, split$test)
#> 
#> ── <PredictionFcst> for 29 observations: ───────────────────────────────────────
#>       month row_ids truth response
#>  1958-08-01     116   505 391.9375
#>  1958-09-01     117   404 324.7500
#>  1958-10-01     118   359 306.0000
#>         ---     ---   ---      ---
#>  1960-10-01     142   461 368.6875
#>  1960-11-01     143   390 364.9333
#>  1960-12-01     144   432 362.3571