Holt-Winters Forecast Learner

Holt-Winters exponential smoothing with optional trend and additive or multiplicative seasonal component. Smoothing parameters are estimated by minimizing the squared one-step prediction error. Calls stats::HoltWinters() from package stats and forecasts via forecast::forecast().

Setting beta = FALSE fits a simple exponential smoothing model (no trend). Setting gamma = FALSE fits a non-seasonal model.

Dictionary

This mlr3::Learner can be instantiated via the dictionary mlr3::mlr_learners or with the associated sugar function mlr3::lrn():

mlr_learners$get("fcst.holt_winters")
lrn("fcst.holt_winters")

Meta Information

• Task type: “fcst”

• Predict Types: “response”, “quantiles”

• Feature Types: “logical”, “integer”, “numeric”, “character”, “factor”, “ordered”, “POSIXct”, “Date”

• Required Packages: mlr3, mlr3forecast, forecast

Parameters

Id	Type	Default	Levels	Range
alpha	numeric	NULL		\([0, 1]\)
beta	numeric	NULL		\([0, 1]\)
gamma	numeric	NULL		\([0, 1]\)
seasonal	character	additive	additive, multiplicative	-
start.periods	integer	2		\([2, \infty)\)
l.start	numeric	NULL		\((-\infty, \infty)\)
b.start	numeric	NULL		\((-\infty, \infty)\)
s.start	untyped	NULL		-
optim.start	untyped	c(alpha = 0.3, beta = 0.1, gamma = 0.1)		-
optim.control	untyped	list()		-
lambda	untyped	NULL		-
biasadj	logical	FALSE	TRUE, FALSE	-

References

Holt, C. C (2004). “Forecasting seasonals and trends by exponentially weighted moving averages.” International Journal of Forecasting, 20(1), 5–10. doi:10.1016/j.ijforecast.2003.09.015 .

Winters, R. P (1960). “Forecasting Sales by Exponentially Weighted Moving Averages.” Management Science, 6(3), 324–342. doi:10.1287/mnsc.6.3.324 .

See also

• Chapter in the mlr3book: https://mlr3book.mlr-org.com/chapters/chapter2/data_and_basic_modeling.html#sec-learners

• Package mlr3learners for a solid collection of essential learners.

• Package mlr3extralearners for more learners.

• Dictionary of Learners: mlr3::mlr_learners

• as.data.table(mlr_learners) for a table of available Learners in the running session (depending on the loaded packages).

• mlr3pipelines to combine learners with pre- and postprocessing steps.

• Package mlr3viz for some generic visualizations.

• Extension packages for additional task types:

  • mlr3proba for probabilistic supervised regression and survival analysis.

  • mlr3cluster for unsupervised clustering.

• mlr3tuning for tuning of hyperparameters, mlr3tuningspaces for established default tuning spaces.

Other Learner: LearnerFcst, mlr_learners_fcst.adam, mlr_learners_fcst.arfima, mlr_learners_fcst.arima, mlr_learners_fcst.auto_adam, mlr_learners_fcst.auto_arima, mlr_learners_fcst.auto_ces, mlr_learners_fcst.auto_gum, mlr_learners_fcst.auto_msarima, mlr_learners_fcst.bagged, mlr_learners_fcst.bats, mlr_learners_fcst.ces, mlr_learners_fcst.croston, mlr_learners_fcst.ets, mlr_learners_fcst.gum, mlr_learners_fcst.mean, mlr_learners_fcst.msarima, mlr_learners_fcst.nnetar, mlr_learners_fcst.prophet, mlr_learners_fcst.random_walk, mlr_learners_fcst.sma, mlr_learners_fcst.spline, mlr_learners_fcst.stlm, mlr_learners_fcst.struct_ts, mlr_learners_fcst.tbats, mlr_learners_fcst.theta, mlr_learners_fcst.tscount, mlr_learners_fcst.tslm

Super classes

mlr3::Learner -> mlr3::LearnerRegr -> LearnerFcst -> LearnerFcstForecast -> LearnerFcstHoltWinters

Methods

Public methods

• LearnerFcstHoltWinters$new()

• LearnerFcstHoltWinters$clone()

Inherited methods

• mlr3::Learner$base_learner()
• mlr3::Learner$configure()
• mlr3::Learner$encapsulate()
• mlr3::Learner$format()
• mlr3::Learner$help()
• mlr3::Learner$predict()
• mlr3::Learner$predict_newdata()
• mlr3::Learner$print()
• mlr3::Learner$reset()
• mlr3::Learner$selected_features()
• mlr3::Learner$train()
• mlr3::LearnerRegr$predict_newdata_fast()

---

LearnerFcstHoltWinters$new()

Creates a new instance of this R6 class.

Usage

LearnerFcstHoltWinters$new()

---

LearnerFcstHoltWinters$clone()

The objects of this class are cloneable with this method.

Usage

LearnerFcstHoltWinters$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# Define the Learner and set parameter values
learner = lrn("fcst.holt_winters")
print(learner)
#> 
#> ── <LearnerFcstHoltWinters> (fcst.holt_winters): Holt-Winters ──────────────────
#> • Model: -
#> • Parameters: list()
#> • Packages: mlr3, mlr3forecast, and forecast
#> • Predict Types: [response] and quantiles
#> • Feature Types: logical, integer, numeric, character, factor, ordered,
#> POSIXct, and Date
#> • Encapsulation: none (fallback: -)
#> • Properties: featureless and missings
#> • Other settings: use_weights = 'error', predict_raw = 'FALSE'

# Define a Task
task = tsk("airpassengers")

# Create train and test set
ids = partition(task)

# Train the learner on the training ids
learner$train(task, row_ids = ids$train)

# Print the model
print(learner$model)
#> Holt-Winters exponential smoothing with trend and additive seasonal component.
#> 
#> Call:
#> stats::HoltWinters(x = as.ts(task))
#> 
#> Smoothing parameters:
#>  alpha: 0.2287362
#>  beta : 0.05161695
#>  gamma: 1
#> 
#> Coefficients:
#>           [,1]
#> a   333.398461
#> b     2.924446
#> s1  -14.893284
#> s2  -25.307089
#> s3   10.725122
#> s4    4.590948
#> s5    6.911731
#> s6   57.136965
#> s7   91.836674
#> s8   76.991025
#> s9   23.706032
#> s10 -26.033269
#> s11 -62.070036
#> s12 -27.398461

# Importance method
if ("importance" %in% learner$properties) print(learner$importance())

# Make predictions for the test rows
predictions = learner$predict(task, row_ids = ids$test)

# Score the predictions
predictions$score()
#> regr.mse 
#> 678.3176