CEGO (version 2.4.3)

optimMIES: Mixed Integer Evolution Strategy (MIES)

Description

An optimization algorithm from the family of Evolution Strategies, designed to optimize mixed-integer problems: The search space is composed of continuous (real-valued) parameters, ordinal integers and categorical parameters. Please note that the categorical parameters need to be coded as integers (type should not be a factor or character). It is an implementation (with a slight modification) of MIES as described by Li et al. (2013). Note, that this algorithm always has a step size for each solution parameter, unlike Li et al., we did not include the option to change to a single step-size for all parameters. Dominant recombination is used for solution parameters (the search space parameters), intermediate recombination for strategy parameters (i.e., step sizes). Mutation: Self-adaptive, step sizes sigma are optimized alongside the solution parameters. Real-valued parameters are subject to variation based on independent normal distributed random variables. Ordinal integers are subject to variation based on the difference of geometric distributions. Categorical parameters are changed at random, with a self-adapted probability. Note, that a more simple bound constraint method is used. Instead of the Transformation Ta,b(x) described by Li et al., optimMIES simply replaces any value that exceeds the bounds by respective boundary value.

Usage

optimMIES(x = NULL, fun, control = list())

Value

a list:

xbest

best solution found.

ybest

fitness of the best solution.

x

history of all evaluated solutions.

y

corresponding target function values f(x).

count

number of performed target function evaluations.

message

Termination message: Which stopping criterion was reached.

population

Last population.

fitness

Fitness of last population.

Arguments

x

Optional start individual(s) as a list. If NULL (default), creationFunction (in control list) is used to create initial design. If x has less individuals than the population size, creationFunction will fill up the rest.

fun

target function to be minimized.

control

(list), with the options:

budget

The limit on number of target function evaluations (stopping criterion) (default: 1000).

popsize

Population size (default: 100).

generations

Number of generations (stopping criterion) (default: Inf).

targetY

Target function value (stopping criterion) (default: -Inf).

vectorized

Boolean. Defines whether target function is vectorized (takes a list of solutions as argument) or not (takes single solution as argument). Default: FALSE.

verbosity

Level of text output during run. Defaults to 0, no output.

plotting

Plot optimization progress during run (TRUE) or not (FALSE). Default is FALSE.

archive

Whether to keep all candidate solutions and their fitness in an archive (TRUE) or not (FALSE). Default is TRUE.

stoppingCriterionFunction

Custom additional stopping criterion. Function evaluated on the population, receiving all individuals (list) and their fitness (vector). If the result is FALSE, the algorithm stops.

types

A vector that specifies the data type of each variable: "numeric", "integer" or "factor".

lower

Lower bound of each variable. Factor variables can have the lower bound set to NA.

upper

Upper bound of each variable. Factor variables can have the upper bound set to NA.

levels

List of levels for each variable (only relevant for categorical variables). Should be a vector of numerical values, usually integers, but not necessarily a sequence. HAS to be given if any factors/categoricals are present. Else, set to NA.

Details

The control variables types, lower, upper and levels are especially important.

References

Rui Li, Michael T. M. Emmerich, Jeroen Eggermont, Thomas Baeck, Martin Schuetz, Jouke Dijkstra, and Johan H. C. Reiber. 2013. Mixed integer evolution strategies for parameter optimization. Evol. Comput. 21, 1 (March 2013), 29-64.

See Also

optimCEGO, optimRS, optimEA, optim2Opt, optimMaxMinDist

Examples

set.seed(1)
controlList <- list(lower=c(-5,-5,1,1,NA,NA),upper=c(10,5,10,10,NA,NA),
		types=c("numeric","numeric","integer","integer","factor","factor"),
	levels=list(NA,NA,NA,NA,c(1,3,5),1:4),
		vectorized = FALSE)
objFun <- function(x){		
		x[[3]] <- round(x[[3]])
		x[[4]] <- round(x[[4]])
		y <- sum(as.numeric(x[1:4])^2) 
		if(x[[5]]==1 & x[[6]]==4)
			y <- exp(y)
		else
			y <- y^2
		if(x[[5]]==3)
			y<-y-1	
		if(x[[5]]==5)
			y<-y-2	
		if(x[[6]]==1)
			y<-y*2
		if(x[[6]]==2)
			y<-y * 1.54
		if(x[[6]]==3)
			y<- y +2
		if(x[[6]]==4)
			y<- y * 0.5	
		if(x[[5]]==1)
			y<- y * 9	
		y	
	}
res <- optimMIES(,objFun,controlList)
res$xbest
res$ybest