DICe::Simplex Class Reference

Non-gradient based optimization algorithm class. More...

#include <DICe_Simplex.h>

Inheritance diagram for DICe::Simplex:

Public Member Functions
	Simplex (const Teuchos::RCP< Teuchos::ParameterList > &params=Teuchos::null)
	Default constructor. More...
virtual	~Simplex ()
	destructor
virtual Status_Flag	minimize (Teuchos::RCP< std::vector< scalar_t > > variables, Teuchos::RCP< std::vector< scalar_t > > deltas, int_t &num_iterations, const scalar_t &threshold=1.0E-10)
	Returns the status of the algorithm when complete. More...
virtual scalar_t	objective (Teuchos::RCP< std::vector< scalar_t > > variables)=0
	the objective function that the simplex method is optimizing More...

Protected Attributes
int_t	max_iterations_
	Maximum allowed iterations for convergence.
double	tolerance_
	Convergence tolerance.
scalar_t	tiny_
	Numerically small value.

Detailed Description

Non-gradient based optimization algorithm class.

Image gradients are not needed or used in a simplex method. The simplex method determines on an optimal solution (one that minimizes the objective's gamma function) by evaluating gamma at simplex vertices and modifying the simplex until it converges to a single point. This method is a lot slower than a gradient-based approach, but is much more robust. The simplex method will almost always converge to a value that that represents the best solution given the input parameters.

Constructor & Destructor Documentation

Simplex()

DICe::Simplex::Simplex

(

const Teuchos::RCP< Teuchos::ParameterList > &

params = Teuchos::null

)

Default constructor.

Parameters

params

Paramters that define the varaitions on the initial guess, convergence tolerance and max number of iterations

Member Function Documentation

minimize()

virtual Status_Flag DICe::Simplex::minimize ( Teuchos::RCP< std::vector< scalar_t > >  variables, Teuchos::RCP< std::vector< scalar_t > >  deltas, int_t &  num_iterations, const scalar_t &  threshold = 1.0E-10  )

virtual

Returns the status of the algorithm when complete.

Parameters

variables	[out] Taken as the initial guess for the first iteration and returned as the converged solution at successful completion.
deltas	The variations on the initial guess used to construct the other simplex points.
num_iterations	[out] Returns the number of iterations that took place
threshold	if the initial evaluation of gamma is below this value, the analysis will be skipped A return value of MAX_ITERATIONS_REACHED means that convergence did not occur in the allowed number of iterations. A return value of CORRELATION_SUCCESSFUL means that convergence was obtained for the solution stored in the variables vector

objective()

virtual scalar_t DICe::Simplex::objective ( Teuchos::RCP< std::vector< scalar_t > >  variables )

pure virtual

the objective function that the simplex method is optimizing

Parameters

variables

the current guess at which to evaluate the objective

Implemented in DICe::Warp_Simplex, DICe::Quadratic_Homography_Simplex, DICe::Affine_Homography_Simplex, DICe::Homography_Simplex, and DICe::Subset_Simplex.