C# Class SwarmOps.Problems.Mangler

Search-space mangler, used to increase the difficulty of optimizing benchmark problems and avoid correlation with global optima such as zero. Note that this works with parallel optimizers but not with parallel meta-optimization because of the way MetaFitness is implemented.

Inheritance: ProblemWrapper

ファイルを表示 Open project: DanWBR/dwsim3

Public Methods

Method	Description
BeginOptimizationRun ( ) : void	At beginning of new optimization run create a random mangling.
Fitness ( double parameters, double fitnessLimit, bool oldFeasible, bool newFeasible ) : double	Compute fitness of wrapped problem and print the result.
Gradient ( double x, double &v ) : int	Compute the gradient of the fitness-function. This is not defined for mangled search-spaces.
Mangler ( Problem problem, double diffusion, double displacement, double spillover, double fitnessNoise ) : System	Constructs a new object.

Method Details

BeginOptimizationRun() public method

At beginning of new optimization run create a random mangling.

public BeginOptimizationRun ( ) : void
return	void

Fitness() public method

Compute fitness of wrapped problem and print the result.

public Fitness ( double parameters, double fitnessLimit, bool oldFeasible, bool newFeasible ) : double
parameters	double
fitnessLimit	double
oldFeasible	bool
newFeasible	bool
return	double

Gradient() public method

Compute the gradient of the fitness-function. This is not defined for mangled search-spaces.

public Gradient ( double x, double &v ) : int
x	double	Candidate solution.
v	double	Array for holding the gradient.
return	int

Mangler() public method

Constructs a new object.

public Mangler ( Problem problem, double diffusion, double displacement, double spillover, double fitnessNoise ) : System
problem	Problem	The problem being wrapped.
diffusion	double	Diffusion factor, larger than 0, e.g. 0.01
displacement	double	Displacement factor, larger than 0, e.g. 0.1
spillover	double	Spillover factor, larger than 0, e.g. 0.05
fitnessNoise	double	FitnessNoise factor, lager than 0, e.g. 0.01
return	System