define_optimizer: Specify numerical optimizer

In optimizeR: Unified Framework for Numerical Optimizers

Specify numerical optimizer

Description

This function specifies the framework for a numerical optimizer.

Two wrappers for well-known optimizers are already available:

1. optimizer_nlm() for the nlm optimizer

2. optimizer_optim() for the optim optimizer

Usage

define_optimizer(
  .optimizer,
  .objective,
  .initial,
  .value,
  .parameter,
  .direction,
  ...,
  .output_ignore = character(0),
  .validate = FALSE,
  .validation_settings = list(objective_test = TestFunctions::TF_ackley, objective_add =
    list(), initial = round(stats::rnorm(2), 2), check_seconds = 10)
)

optimizer_nlm(
  ...,
  .output_ignore = character(0),
  .validate = FALSE,
  .validation_settings = list()
)

optimizer_optim(
  ...,
  .direction = "min",
  .output_ignore = character(0),
  .validate = FALSE,
  .validation_settings = list()
)

Arguments

.optimizer	A function, a numerical optimizer. Four conditions must be met: It must have an input named .objective for a function, the objective function which is optimized over its first argument. It must have an input named .initial for a numerical vector, the initial parameter vector. It must have a ... argument for additional parameters to the objective function. The output must be a named list, including the optimal function value and the optimal parameter vector.
.objective	A character, the name of the function input of optimizer.
.initial	A character, the name of the starting parameter values input of optimizer.
.value	A character, the name of the optimal function value in the output list of optimizer.
.parameter	A character, the name of the optimal parameter vector in the output list of optimizer.
.direction	A character, indicates whether the optimizer minimizes ("min") or maximizes ("max").
...	Additional arguments to be passed to the optimizer. Without specifications, the default values of the optimizer are used.
.output_ignore	A character vector of element names in the output of .optimizer that are not saved. The elements .value and .parameter are added automatically to .output_ignore, because they are saved separately, see the output documentation of apply_optimizer.
.validate	A logical, set to TRUE (FALSE) to (not) validate the optimizer object. By default, .validate = FALSE.
.validation_settings	Ignored if .valdiate = FALSE. Otherwise, a list of validation settings: objective_test A function, the test function to be optimized. By default, it is the Ackley function. objective_add A list of additional arguments to objective_test (if any). By default, objective_add = list(), because the default function for objective_test does not have additional arguments. initial A numeric vector, the initial values for the optimization of objective_test. By default, initial = round(stats::rnorm(2), 2). check_seconds An integer, the maximum number of seconds before the test is aborted. The test call is considered to be successful if no error occurred within check_seconds seconds. By default, check_seconds = 10.

Value

An optimizer object.

Format

An optimizer object is a list of six elements:

optimizer

A function, the optimization algorithm.

optimizer_name

A character, the name of optimizer.

optimizer_arguments

A named list, where each element is an additional function argument for optimizer.

optimizer_direction

Either "min" if the optimizer minimizes or "max" if the optimizer maximizes.

optimizer_labels

A named list of four character:

objective

the name of the function input of optimizer

initial

the name of the starting parameter values input of optimizer

value

the name of the optimal function value in the output list of optimizer

parameter

the name of the optimal parameter vector in the output list of optimizer.

output_ignore

A character vector of element names in the output list of optimizer that are ignored. The elements value and parameter are added automatically to output_ignore, because they are saved separately, see the output documentation of apply_optimizer.

See Also

Use apply_optimizer() to apply an optimizer object for numerical optimization.

Examples

define_optimizer(
  .optimizer = pracma::nelder_mead,           # optimization function
  .objective = "fn",                          # name of function input
  .initial = "x0",                            # name of initial input
  .value = "fmin",                            # name of value output
  .parameter = "xmin",                        # name of parameter output
  .direction = "min",                         # optimizer minimizes
  .output_ignore = c("restarts", "errmess"),  # ignore some outputs
  tol = 1e-6,                                 # additional optimizer argument
  .validate = TRUE                            # validate the object
)

optimizeR documentation built on April 4, 2025, 1:56 a.m.