Skip to contents

Create Rolling Window Features of Target Variable

Source: R/PipeOpFcstRolling.R
mlr_pipeops_fcst.rolling.Rd

Creates rolling-window summary statistics of the target variable as new feature columns. The window ends at position t - lag (exclusive of the current and lag - 1 most recent values) and has size window_size. Use window_size = Inf for an expanding window that grows to include all history up to t - lag.

At train time rows whose window has insufficient history are NA and are dropped, matching PipeOpFcstLags. Predict keeps all rows.

At predict time, rolling features are computed from the task's full backend (i.e. including rows outside row_roles$use), then joined onto the active rows. Used inside RecursiveForecaster, where the forecaster writes each step's prediction into the combined task's target column between steps so rolling features for the next step reflect the freshly predicted value.

Parameters

The parameters are the parameters inherited from mlr3pipelines::PipeOpTaskPreproc, as well as the following parameters:

  • funs :: character()
    Aggregation functions. Subset of c("mean", "median", "sd", "min", "max", "sum"). Default "mean".

  • window_sizes :: numeric()
    Window sizes. Every combination of funs and window_sizes produces one output column. Finite sizes must be whole numbers. Inf requests an expanding window (all history up to t - lag). Default 3L.

  • lag :: integer(1)
    Minimum lag before the window starts. Must be >= 1 to avoid leakage. Default 1L.

Super classes

mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpTaskPreproc -> PipeOpFcstRolling

Methods

Public methods

Inherited methods

PipeOpFcstRolling$new()

Initializes a new instance of this Class.

Usage

PipeOpFcstRolling$new(id = "fcst.rolling", param_vals = list())

Arguments

id

(character(1))
Identifier of resulting object, default "fcst.rolling".

param_vals

(named list())
List of hyperparameter settings, overwriting the hyperparameter settings that would otherwise be set during construction. Default list().

PipeOpFcstRolling$clone()

The objects of this class are cloneable with this method.

Usage

PipeOpFcstRolling$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

library(mlr3pipelines)
task = tsk("airpassengers")
po = po("fcst.rolling", funs = c("mean", "sd"), window_sizes = c(3L, 12L))
new_task = po$train(list(task))[[1L]]
new_task$head()
#>    passengers passengers_roll_mean_3 passengers_roll_mean_12 passengers_roll_sd_3 passengers_roll_sd_12
#>         <num>                  <num>                   <num>                <num>                 <num>
#> 1:        115               113.6667                126.6667             8.386497              13.72015
#> 2:        126               112.3333                126.9167             7.371115              13.45334
#> 3:        141               119.6667                127.5833             5.686241              13.16647
#> 4:        135               127.3333                128.3333            13.051181              13.68698
#> 5:        125               134.0000                128.8333             7.549834              13.82247
#> 6:        149               133.6667                129.1667             8.082904              13.66371