mlr3mbo (version 1.2.1)

AcqOptimizerLbfgsb: L-BFGS-B Acquisition Function Optimizer

Description

L-BFGS-B acquisition function optimizer. Calls nloptr() from nloptr with the NLOPT_LD_LBFGS algorithm. In its default setting, the algorithm runs a single time starting from the best point in the archive, for at most 100 * D^2 function evaluations, where D is the dimension of the search space. The run stops when the relative tolerance of the parameters is less than 10^-4. With restart_strategy = "random", the optimizer additionally restarts from random points until the evaluation budget is exhausted, which can help escape local optima.

Only fully numeric search spaces (all parameters of type p_dbl) are supported.

Parameters

restart_strategy

character(1)
Restart strategy. Can be "none" or "random". Default is "none".

max_restarts

integer(1)
Maximum number of restarts. Default is 5 * D, where D is the dimension of the search space.

skip_already_evaluated

logical(1)
Should the proposed candidate be rejected if it was already evaluated on the actual bbotk::OptimInstance? If TRUE and the candidate was already evaluated, an error is raised so that the loop_function can propose a randomly sampled point instead. Default is TRUE.

Termination Parameters

The following termination parameters can be used.

stopval

numeric(1)
Stop value. Deactivate with -Inf (Default).

maxeval

integer(1)
Maximum number of evaluations. Default is 100 * D^2, where D is the dimension of the search space. Deactivate with -1L.

xtol_rel

numeric(1)
Relative tolerance of the parameters. Default is 10^-4. Deactivate with -1.

xtol_abs

numeric(1)
Absolute tolerance of the parameters. Deactivate with -1 (Default).

ftol_rel

numeric(1)
Relative tolerance of the objective function. Deactivate with -1 (Default).

ftol_abs

numeric(1)
Absolute tolerance of the objective function. Deactivate with -1 (Default).

Super class

AcqOptimizer -> AcqOptimizerLbfgsb

Public fields

state

(list())
List of nloptr::nloptr() results.

Active bindings

print_id

(character)
Id used when printing.

label

(character(1))
Label for this object. Can be used in tables, plot and text output instead of the ID.

man

(character(1))
String in the format [pkg]::[topic] pointing to a manual page for this object.

Methods

Public methods

• AcqOptimizerLbfgsb$new()

• AcqOptimizerLbfgsb$optimize()

• AcqOptimizerLbfgsb$reset()

• AcqOptimizerLbfgsb$clone()

Inherited methods

• AcqOptimizer$format()
• AcqOptimizer$print()

---

AcqOptimizerLbfgsb$new()

Creates a new instance of this R6 class.

Usage

AcqOptimizerLbfgsb$new(acq_function = NULL)

Arguments

acq_function

(NULL | AcqFunction).

---

AcqOptimizerLbfgsb$optimize()

Optimize the acquisition function.

Usage

AcqOptimizerLbfgsb$optimize()

Returns

data.table::data.table() with 1 row per candidate.

---

AcqOptimizerLbfgsb$reset()

Reset the acquisition function optimizer.

Clears the state of the previous optimization run.

Usage

AcqOptimizerLbfgsb$reset()

---

AcqOptimizerLbfgsb$clone()

The objects of this class are cloneable with this method.

Usage

AcqOptimizerLbfgsb$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

if (requireNamespace("nloptr")) {
  acqo("lbfgsb")
}