auto_ets: Automatic Model Selection

In tsmodels/tsets: Exponential Smoothing Models

Automatic Model Selection

Description

Automatic model selection based on choice of criteria.

Usage

auto_ets(
  y,
  xreg = NULL,
  transformation = NULL,
  lambda = NULL,
  lower = 0,
  upper = 1,
  metric = "AIC",
  frequency = NULL,
  normalized_seasonality = TRUE,
  additive_only = FALSE,
  solver = "nlminb",
  control = list(trace = 0, maxit = 1000),
  power_model = FALSE,
  include_damped = TRUE,
  trace = FALSE,
  return_table = FALSE,
  scale = FALSE,
  seasonal_init = "fixed",
  autodiff = TRUE,
  ...
)

Arguments

y	an xts vector.
xreg	an optional xts matrix of regressors (pre-lagged).
transformation	a valid transformation for y from the “tstransform” function in the “tsaux” package (currently box-cox or logit are available) applied to additive models only.
lambda	the Box Cox power parameter (lambda). If NA will estimate this using the method of Guerrero.
lower	the lower bound for the transformation.
upper	the upper bound for the transformation.
metric	the selection metric to use. Valid metrics are ‘AIC’, ‘BIC’, ‘AICc’, ‘MASE’ and ‘MAPE’. If lambda is not NULL, then those models which admit a Box-Cox transformation (additive error models) will not be comparable with the other models.
frequency	the frequency of y (if using a seasonal model).
normalized_seasonality	whether to impose Roberts-McKenzie normalized seasonality.
additive_only	whether to limit to additive models only.
solver	the solver to use for estimation.
control	the solver control parameters.
power_model	whether to include the power MAM models.
include_damped	whether to include damped models in the selection.
trace	whether to show the progress bar. The user is expected to have set up appropriate handlers for this using the “progressr” package.
return_table	whether to return the table with the enumerated options, ranked by metric,for each combination of those options used.
scale	whether to rescale the data using y/max(y) (only for additive models). This sometimes helps in the optimization.
seasonal_init	whether the initial seasonal states are estimated or fixed (set to a backcast approximation).
autodiff	whether to use automatic differentiation (see estimate.tsets.spec).
...	not used.

Details

The function estimates a number of different models from the class of valid ETS models in the packages, with and without dampening, with and without a power term (for MAM and MAN models).

Value

An object of class “tsets.estimate” which also inherits class “tsets.select”

Note

The function can use parallel functionality as long as the user has set up a plan using the future package.

tsmodels/tsets documentation built on Oct. 8, 2022, 9:15 a.m.