General Interface for Exponential Smoothing Models

exponential_smoothing() is a way to generate a specification of an ETS model before fitting and allows the model to be created using different packages. Currently the only package is Rlgt.

exponential_smoothing(
  mode = "regression",
  seasonality = NULL,
  second_seasonality = NULL,
  seasonality_type = NULL,
  method = NULL,
  error_method = NULL
)

Arguments

mode	A single character string for the type of model. The only possible value for this model is "regression".
seasonality	This specification of seasonality will be overridden by frequency of y, if y is of ts or msts class. 1 by default, i.e. no seasonality.
second_seasonality	Second seasonality.
seasonality_type	Either "multiplicative" (default) or "generalized". The latter seasonality generalizes additive and multiplicative seasonality types.
method	"HW", "seasAvg", "HW_sAvg". Here, "HW" follows Holt-Winters approach. "seasAvg" calculates level as a smoothed average of the last seasonality number of points (or seasonality2 of them for the dual seasonality model), and HW_sAvg is an weighted average of HW and seasAvg methods.
error_method	Function providing size of the error. Either "std" (monotonically, but slower than proportionally, growing with the series values) or "innov" (proportional to a smoothed abs size of innovations, i.e. surprises)

Value

A model spec

Details

The data given to the function are not saved and are only used to determine the mode of the model. For exponential_smoothing(), the mode will always be "regression".

The model can be created using the fit() function using the following engines:

• "stan" (default) - Connects to Rlgt::rlgt()

Main Arguments

The main arguments (tuning parameters) for the model are:

• seasonality: Seasonality.

• second_seasonality: Second seasonality.

• seasonality_type: Either "multiplicative" (default) or "generalized".

• method: "HW", "seasAvg", "HW_sAvg"

• error_method: Either "std" or "innov"

These arguments are converted to their specific names at the time that the model is fit.

Other options and argument can be set using set_engine().

If parameters need to be modified, update() can be used in lieu of recreating the object from scratch.

stan (default engine)

The engine uses Rlgt::rlgt().

Parameter Notes:

• xreg - This is supplied via the parsnip / bayesmodels fit() interface (so don't provide this manually). See Fit Details (below).

Fit Details

Date and Date-Time Variable

It's a requirement to have a date or date-time variable as a predictor. The fit() interface accepts date and date-time features and handles them internally.

• fit(y ~ date)

Univariate (No xregs, Exogenous Regressors):

For univariate analysis, you must include a date or date-time feature. Simply use:

• Formula Interface: fit(y ~ date) will ignore xreg's.

Multivariate (xregs, Exogenous Regressors)

The xreg parameter is populated using the fit() function:

• Only factor, ordered factor, and numeric data will be used as xregs.

• Date and Date-time variables are not used as xregs

• character data should be converted to factor.

Xreg Example: Suppose you have 3 features:

1. y (target)

2. date (time stamp),

3. month.lbl (labeled month as a ordered factor).

The month.lbl is an exogenous regressor that can be passed to the expotential_smoothing() using fit():

• fit(y ~ date + month.lbl) will pass month.lbl on as an exogenous regressor.

Note that date or date-time class values are excluded from xreg.

See also

fit.model_spec(), set_engine()

Examples

if (FALSE) {
library(dplyr)
library(parsnip)
library(rsample)
library(timetk)
library(modeltime)
library(bayesmodels)

# Data
m750 <- m4_monthly %>% filter(id == "M750")
m750

# Split Data 80/20
splits <- rsample::initial_time_split(m750, prop = 0.8)

# ---- ARIMA ----

# Model Spec
model_spec <- exponential_smoothing() %>%
    set_engine("stan")

# Fit Spec
model_fit <- model_spec %>%
    fit(log(value) ~ date + month(date), data = training(splits))
model_fit
}