cmaEvalMeanX() R function from [rCMA]

Evaluate the meanX of the current population.

After executing cmaOptimDP, there is a current population and a best-ever solution. Evaluate for the mean of the current population whether it is feasible and whether the mean is an even better solution. If so, update the best-ever solution.


cmaEvalMeanX(cma, fitFunc, isFeasible = function(x) TRUE)

Arguments

cma: CMA-ES Java object, already initialized with cmaInit
fitFunc: a function to be minimized. Signature: accepts a vector x, returns a double.
isFeasible: [function(x){TRUE}] a Boolean function checking the feasibility of the vector x. The default is to return always TRUE.

Returns

bestSolution, a list with entries: - bestX: a vector of length dimension containing the best-ever solution, including meanX

meanX: a vector of length dimension containing the mean of the current (last) population in cma
bestFitness: the best-ever fitness value, including the evaluation of meanX
bestEvalNum: the function evaluation count where bestFitness occured
lastEvalNum: the total function evaluation count. If bestEvalNum==lastEvalNum

then the best-ever fitness occured in the evaluation of meanX.

Details

The code of this function is also instructive as a full example for the extensibility of the list("rJava")

interface to CMA-ES. See the full code in demo/demoEvalMeanX. Some example rJava-calls are:

rJava::.jcall(cma,"[D","getMeanX");
   bestSolutionObj =
rJava::.jcall(cma,"Lfr/inria/optimization/cmaes/CMASolution;","setFitnessOfMeanX",fitFunc(meanX));
   rJava::.jcall(bestSolutionObj,"J","getEvaluationNumber");

Every direct method of classes in the CMA-ES Java package cmaes (see [Hansen09] for the complete Javadoc and [Hansen13] for an overview on CMA-ES in total) can be accessed with the .jcall-mechanism of the list("rJava") R package:

rJava::.jcall(obj,returnType,method,...)

where ... stands for the calling parameter(s) of method.

returnType is a string following the JNI type convention (see, e.g. [Oracle14])


	Field Descriptor	Java Language Type
	Z	boolean
	C	char
	I	int
	J	long
	F	float
	D	double
	[I	int[]
	[[D	double[][]
	Ljava/langString;	java.lang.String
	S	java.lang.String
	T	short

(Note: (a) the terminating ";" in "Ljava/langString;" (!) and (b) "S" is a short hand for "Ljava/langString;" and "T" is the re-mapped code for short. )

The calling parameters in ... have to be matched exactly. In R, numeric vectors are stored as doubles, so the calling syntax

bestSolutionObj = .jcall(cma,rType,"setFitnessOfMeanX",fitFunc(meanX));

is just right for the Java method setFitnessOfMeanX(double[]). In other cases, the calling R variable x

has to be cast explicitly:


	Cast	Java Language Type
	.jbyte(x)	byte
	.jchar(x)	char
	as.integer(x)	int
	.jlong(x)	long
	.jfloat(x)	float

Examples


## Not run:

   ## just to show the syntax, without calling cmaOptimDP
   fitFunc <- function(x) {  sum(x*x); }
   isFeasible <- function(x) {  TRUE;  }
   cma <- cmaNew(propFile="CMAEvolutionStrategy.properties");
   cmaInit(cma,dimension=2,initialX=1.5);
   bestSolution=cmaEvalMeanX(cma,fitFunc,isFeasible);
   str(bestSolution);
## End(Not run)

Author(s)

Wolfgang Konen, FHK, 2013-2015

References

[Hansen09] https://www.lri.fr/~hansen/javadoc Nikolaus Hansen: Javadoc for CMA-ES Java package fr.inria.optimization.cmaes, 2009.

[Hansen13] https://www.lri.fr/~hansen/cmaesintro.html Nikolaus Hansen: The CMA Evolution Strategy, 2013.

[Oracle14] http://docs.oracle.com/javase/7/docs/technotes/guides/jni/spec/jniTOC.html

Oracle: The Java Native Interface. Programmer's Guide and Specification. Chapter 3 (JNI types), Sec. 'Type Signatures', 2014.

cmaEvalMeanX function