GClasses::GSampleClimber Class Reference

This is a variant of empirical gradient descent that tries to estimate the gradient using a minimal number of samples. It is more efficient than empirical gradient descent, but it only works well if the optimization surface is quite locally linear. More...

#include <GHillClimber.h>

Inheritance diagram for GClasses::GSampleClimber:

Public Member Functions
	GSampleClimber (GTargetFunction *pCritic, GRand *pRand)
virtual	~GSampleClimber ()
virtual double *	currentVector ()
	Returns the best vector yet found. More...
virtual double	iterate ()
	Performs a little more optimization. (Call this in a loop until acceptable results are found.) More...
void	setAlpha (double d)
	Sets the alpha value. It should be small (like 0.01) A very small value updates the gradient estimate slowly, but precisely. A bigger value updates the estimate quickly, but never converges very close to the precise gradient. More...
void	setStepSize (double d)
	Sets the current step size. More...
Public Member Functions inherited from GClasses::GOptimizer
	GOptimizer (GTargetFunction *pCritic)
virtual	~GOptimizer ()
double	searchUntil (size_t nBurnInIterations, size_t nIterations, double dImprovement)
	This will first call iterate() nBurnInIterations times, then it will repeatedly call iterate() in blocks of nIterations times. If the error heuristic has not improved by the specified ratio after a block of iterations, it will stop. (For example, if the error before the block of iterations was 50, and the error after is 49, then training will stop if dImprovement is > 0.02.) If the error heuristic is not stable, then the value of nIterations should be large. More...

Protected Member Functions
void	reset ()

Protected Attributes
double	m_alpha
size_t	m_dims
double	m_dStepSize
double	m_error
double *	m_pCand
double *	m_pDir
double *	m_pGradient
GRand *	m_pRand
double *	m_pVector
Protected Attributes inherited from GClasses::GOptimizer
GTargetFunction *	m_pCritic

Detailed Description

This is a variant of empirical gradient descent that tries to estimate the gradient using a minimal number of samples. It is more efficient than empirical gradient descent, but it only works well if the optimization surface is quite locally linear.

Constructor & Destructor Documentation

GClasses::GSampleClimber::GSampleClimber	(	GTargetFunction *	pCritic,
		GRand *	pRand
	)

virtual GClasses::GSampleClimber::~GSampleClimber ( )

virtual

Member Function Documentation

virtual double* GClasses::GSampleClimber::currentVector ( )

inlinevirtual

Returns the best vector yet found.

Implements GClasses::GOptimizer.

virtual double GClasses::GSampleClimber::iterate ( )

virtual

Performs a little more optimization. (Call this in a loop until acceptable results are found.)

Implements GClasses::GOptimizer.

void GClasses::GSampleClimber::reset ( )

protected

void GClasses::GSampleClimber::setAlpha ( double  d )

inline

Sets the alpha value. It should be small (like 0.01) A very small value updates the gradient estimate slowly, but precisely. A bigger value updates the estimate quickly, but never converges very close to the precise gradient.

void GClasses::GSampleClimber::setStepSize ( double  d )

inline

Sets the current step size.

Member Data Documentation

double GClasses::GSampleClimber::m_alpha

protected

size_t GClasses::GSampleClimber::m_dims

protected

double GClasses::GSampleClimber::m_dStepSize

protected

double GClasses::GSampleClimber::m_error

protected

double* GClasses::GSampleClimber::m_pCand

protected

double* GClasses::GSampleClimber::m_pDir

protected

double* GClasses::GSampleClimber::m_pGradient

protected

GRand* GClasses::GSampleClimber::m_pRand

protected

double* GClasses::GSampleClimber::m_pVector

protected