compass< LOSSFUNC, PARAM, BOX > Struct Template Reference

Compass Search. More...

#include <compass.h>

Public Member Functions
	BBM_IMPORT_CONFIG (LOSSFUNC)
	compass (LOSSFUNC &lossfunc, PARAM &param, const BOX &lower=BOX(), const BOX &upper=BOX(), Scalar tolerance=Constants::Epsilon(), Scalar stepSize=1.0, Scalar contraction=0.5, Scalar expansion=1.0, Mask mask=true)
	Constructor: compass search.
Value	step (void)
	probe each cardinal direction, and update the parameters to the one with the lowest loss.
void	reset (void)
	reset the step size
Mask	is_converged (void) const
	is_converged when _step < _tolerance

Private Attributes
LOSSFUNC &	_lossfunc
PARAM &	_param
BOX	_lower
BOX	_upper
Value	_step
Value	_lossValue
std::vector< Scalar >	_directions
Scalar	_initialStep
Scalar	_tolerance
Scalar	_contraction
Scalar	_expansion
Mask	_mask

Detailed Description

template<typename LOSSFUNC, typename PARAM, typename BOX = PARAM>
requires concepts::lossfunction<LOSSFUNC> && concepts::parameter<PARAM> && concepts::parameter<BOX>
struct bbm::compass< LOSSFUNC, PARAM, BOX >

Compass Search.

Template Parameters

LOSSFUNC	= loss function
PARAM	= parameters to optimize
BOX	= box constraint type (default=PARAM)

Performs a pattern search by probing each cardinal direction in the parameter space.

When the Value type is a packet, this compass search will perform N parallel searches (where N == size of the packet).

Satisfies: concepts::optimization algorithm

Constructor & Destructor Documentation

compass()

compass ( LOSSFUNC &  lossfunc, PARAM &  param, const BOX &  lower = BOX(), const BOX &  upper = BOX(), Scalar  tolerance = Constants::Epsilon(), Scalar  stepSize = 1.0, Scalar  contraction = 0.5, Scalar  expansion = 1.0, Mask  mask = true  )

inline

Constructor: compass search.

Parameters

lossfunc	= loss function to minimize
param	= set of parameters to optimize
lower	= lower bound of box constraint
upper	= upper bound of box constraint
tolerance	= threshold on step size to determine convergence (default=Epsilon)
stepSize	= initial step size (default=1.0)
contraction	= reduction factor for step size if no probe in the cardinal directions improves the loss (default=0.5)
expansion	= expansion factor for step size if a good candinal direction was found (default=1.0)

Member Function Documentation

BBM_IMPORT_CONFIG()

BBM_IMPORT_CONFIG

(

LOSSFUNC

)

is_converged()

Mask is_converged ( void  ) const

inline

is_converged when _step < _tolerance

reset()

void reset ( void  )

inline

reset the step size

step()

Value step ( void  )

inline

probe each cardinal direction, and update the parameters to the one with the lowest loss.

Returns

loss after update

Member Data Documentation

_contraction

Scalar _contraction

private

_directions

std::vector<Scalar> _directions

private

_expansion

Scalar _expansion

private

_initialStep

Scalar _initialStep

private

_lossfunc

LOSSFUNC& _lossfunc

private

_lossValue

Value _lossValue

private

_lower

BOX _lower

private

_mask

Mask _mask

private

_param

PARAM& _param

private

_step

Value _step

private

_tolerance

Scalar _tolerance

private

_upper

BOX _upper

private

---

The documentation for this struct was generated from the following file:

• include/optimizer/compass.h