Skip to contents

Objective function

Objective.Rd

This class defines the objective that is used for optimization.

Value

Returns self invisibly.

Public fields

id

(character(1))
The id of the object.

label

(character(1) The label of the objective, i.e. a.

lower

(numeric()) The lower limits for each dimension.

upper

(numeric()) The upper limits for each dimension.

minimize

(logical(1)) Is the problem a minimization problem?

Active bindings

archive

(data.table()) Archive of all calls to $eval_store.

log_funs

(list()) A list containing logging functions. Each function must have argument.

xdim

(integer(1)) Input dimension of f.

transform_id

(character(1)) Identifier of the active transform.

Methods

Public methods

Method new()

Creates a new instance of this R6 class.

Usage

Objective$new(
  id,
  fun,
  label = "f",
  xdim,
  lower = NA,
  upper = NA,
  xtest = NULL,
  minimize = FALSE,
  transform = objective_transform_identity(),
  label_base = NULL,
  ...
)

Arguments

id

(character(1))
Id of the object.

fun

(function The objective function. The first argument must be a numerical input of length xdim.

label

(character(1) The label of the objective, i.e. a.

xdim

(integer(1)) The input dimension of fun. Use xdim = NA for an arbitrary input dimension.

lower

(numeric(xdim)) The lower boundaries for inputs to fun. Must

upper

(numeric(xdim)) The upper boundaries for inputs to fun. Must be of length xdim.

xtest

(numeric()) Test value for fun during initialization. If not defined, xtest = rep(0, ifelse(is.na(xdim), 2, xdim)) is used.

minimize

(logical(1)) Is the problem a minimization problem? Default is no (FALSE).

transform

(objective_transform()) Optional scalar transformation applied to the objective value. Defaults to objective_transform_identity().

label_base

(character(1)) Optional base label used when appending transform identifiers. Defaults to label.

...


Additional arguments passed to fun.

Method eval()

Evaluate the objective function.

Usage

Objective$eval(x)

Arguments

x

(numeric) The numerical input of fun.

Returns

The result of fun(x).

Method eval_store()

Evaluate the objective function and log into the archive. Each call logs the input vector x, result of fun fval, the gradient grad, the norm of the gradient gnorm, and additional logs that were added by $add_log_fun.

Usage

Objective$eval_store(x)

Arguments

x

(numeric) The numerical input of fun.

Returns

Invisible list of logs that are added to the archive.

Method assert_x()

Assert a numeric input if it is suitable or not.

Usage

Objective$assert_x(x, ...)

Arguments

x

(numeric()) Input value for fun.

...

Additional arguments passed to assertNumeric(...).

Method set_transform()

Update the active transformation for the objective.

Usage

Objective$set_transform(transform, test = TRUE)

Arguments

transform

(objective_transform()) Transformation descriptor created via objective_transform().

test

(logical(1)) If TRUE, validates the transform by evaluating the objective at the stored test point.

Returns

The Objective invisibly.

Method get_transform()

Retrieve the active transformation descriptor.

Usage

Objective$get_transform()

Returns

An object created by objective_transform().

Method grad()

Evaluate the gradient of the objective function at x.

Usage

Objective$grad(x)

Arguments

x

(numeric) The numerical input of fun.

Method hess()

Evaluate the hessian of the objective function at x.

Usage

Objective$hess(x)

Arguments

x

(numeric) The numerical input of fun.

Method add_log_fun()

Method to add custom logger to the objective.

Usage

Objective$add_log_fun(l, label)

Arguments

l

(function) Function that returns a single numerical value or a string. The arguments of l must be x, fval and grad.

label

(character(1)) The name of the logger.

Method clear_archive()

Delete the archive.

Usage

Objective$clear_archive()

Method clone()

The objects of this class are cloneable with this method.

Usage

Objective$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.