spotControl: spotControl

Description

Default Control list for spot. This function returns the default controls for the functions spot and spotLoop.

Usage

spotControl(dimension)

Arguments

dimension

problem dimension, that is, the number of optimized parameters.

Value

a list

Details

Control is a list of the settings:

funEvals: This is the budget of function evaluations (spot uses no more than funEvals evaluations of fun), defaults to 20.
types: Vector of data type of each variable as a string, defaults "numeric" for all variables.
subsetSelect: A function that selects a subset from a given set of design points. Default is selectAll.
subsetControl: A list of controls passed to the control list of the subsetSelect function. See help of the respective function for details. Default is an empty list.
design: A function that creates an initial design of experiment. Functions that accept the same parameters, and return a matrix like designLHD or designUniformRandom can be used. Default is designLHD.
designControl: A list of controls passed to the control list of the design function. See help of the respective function for details. Default is an empty list.
model: A function that builds a statistical model of the observed data. Functions that accept the same parameters, and return a matrix like buildKriging or buildRandomForest can be used. Default is buildKriging.
modelControl: A list of controls passed to the control list of the model function. See help of the respective function for details.Default is an empty list.
optimizer: A function that is used to optimize based on model, finding the most promising candidate solutions. Functions that accept the same parameters, and return a matrix like optimLHD or optimDE can be used. Default is optimLHD.
optimizerControl: A list of controls passed to the control list of the optimizer function. See help of the respective function for details. Default is an empty list.
directOpt: A function that is used to optimize after the spot run is finished. Functions that accept the same parameters, and return a matrix like optimNLOPTR or optimDE can be used. Default is optimNLOPTR.
directOptControl: A list of controls passed to the control list of the directOpt function. See help of the respective function for details. Default is list(funEvals = 0).
noise: Boolean, whether the objective function has noise or not. Default is non-noisy, that is, FALSE.
OCBA: Boolean, indicating whether Optimal Computing Budget Allocation (OCBA) should be used in case of a noisy objective function or not. OCBA controls the number of replications for each candidate solution. Note, that replicates should be larger than one in that case, and that the initial experimental design (see design) should also have replicates larger one. Default is FALSE.
OCBAbudget: The number of objective function evaluations that OCBA can distribute in each iteration. Default is 3.
replicates: The number of times a candidate solution is initially evaluated, that is, in the initial design, or when created by the optimizer. Default is 1.
seedFun: An initial seed for the objective function in case of noise, by default NA. The default means that no seed is set. The user should be very careful with this setting. It is intended to generate reproducible experiments for each objective function evaluation, e.g., when tuning non-deterministic algorithms. If the objective function uses a constant number of random number generations, this may be undesirable. Note, that this seed is by default set prior to each evaluation. A replicated evaluation will receive an incremented value of the seed. Sometimes, the user may want to call external code using random numbers. To allow for that case, the user can specify an objective function (fun), which has a second parameter seed, in addition to first parameter (matrix x). This seed can then be passed to the external code, for random number generator initialization. See end of examples section for a demonstration.
seedSPOT: This value is used to initialize the random number generator. It ensures that experiments are reproducible. Default is 1.
duplicate: In case of a deterministic (non-noisy) objective function, this handles duplicated candidate solutions. By default (duplicate = "EXPLORE"), duplicates are replaced by new candidate solutions, generated by random sampling with uniform distribution. If desired, the user can set this to "STOP", which means that the optimization stops and results are returned to the user (with a warning). This may be desirable, as duplicates can be a indicator for convergence, or for a problem with the configuration. In case of noise, duplicates are allowed.
plots: Whether progress should be tracked by a line plot, default is FALSE
progress: Whether progress should be visualized, default is FALSE
infillCriterion: A function defining an infillCriterion to be used while optimizing a model. Default: NULL. For example check infillExpectedImprovement
verbosity: Integer level specifying how much output should be given by SPOT. 0 (default) ignores warnings of internal optimizers /models. 1 will show warnings and output.
maxTime: num Maximum allowed run time (in minutes) for spot or spotLoop. The default value for maxTime (in minutes) is Inf and can be overwritten by the user. The internal value startTime, that is used to control maxTime, will be set by spotFillControlList. Note: maxTime is only an approximate value. It does not affect the directOpt run.