cmp::ObjectiveFunctor Class Reference

Functor wrapper for NLopt with automatic, transparent parameter space mapping (e.g., log-scaling). More...

#include <optimization.h>

Classes
struct	ConstraintContext
	Context struct for NLopt constraint evaluation callback. More...

Public Member Functions
	ObjectiveFunctor (std::function< double(Eigen::Ref< const Eigen::VectorXd >)> fval)
	ObjectiveFunctor (std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> fval_grad)
void	setLogScale (const std::vector< bool > &log_scale)
	Set which parameters should be optimized in log-space.
const std::vector< bool > &	getLogScale () const
double	operator() (Eigen::Ref< const Eigen::VectorXd > x_opt, Eigen::Ref< Eigen::VectorXd > grad_opt) const
bool	usesGradient () const
	Checks if the functor utilizes gradient information.
void	addInequalityConstraint (std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> g)
void	addEqualityConstraint (std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> h)
std::vector< std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> >	getInequalityConstraints () const
std::vector< std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> >	getEqualityConstraints () const

Static Public Member Functions
static double	NLoptCallback (const std::vector< double > &x, std::vector< double > &grad, void *data)
static double	NLoptConstraintWrapper (const std::vector< double > &x, std::vector< double > &grad, void *data)

Private Member Functions
Eigen::VectorXd	mapToReal (Eigen::Ref< const Eigen::VectorXd > x_opt) const
	Maps parameters from optimization space (potentially log-scaled) to physical space.
void	mapGradientToOpt (Eigen::Ref< const Eigen::VectorXd > x_real, Eigen::Ref< Eigen::VectorXd > grad_real) const
	Transforms gradients from real space back to optimization space using the chain rule.
double	evaluateConstraint (size_t index, bool is_ineq, Eigen::Ref< const Eigen::VectorXd > x_opt, Eigen::Ref< Eigen::VectorXd > grad_opt) const
	Evaluates a constraint while properly translating inputs and mapping output gradients.

Private Attributes
std::function< double(Eigen::Ref< const Eigen::VectorXd >)>	fval_only_
	Pointer to a gradient-free objective function.
std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)>	fval_grad_inplace_
	Pointer to a gradient-based objective function.
std::vector< std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> >	ineq_constraints_
	Collection of inequality constraint functions.
std::vector< std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> >	eq_constraints_
	Collection of equality constraint functions.
bool	use_gradient_
	Flag indicating whether the objective function uses gradient information.
std::vector< bool >	log_scale_
	Mask indicating which parameter dimensions are optimized in log-scale.

Detailed Description

Functor wrapper for NLopt with automatic, transparent parameter space mapping (e.g., log-scaling).

Mathematical Formulation When optimization parameters span multiple orders of magnitude, log-scaling maps the optimization coordinate \(y_k\) to the physical parameter \(x_k\):

\[ x_k = \exp(y_k) \]

By the chain rule, the gradient evaluated by NLopt with respect to \(y_k\) is:

\[ \frac{\partial f}{\partial y_k} = \frac{\partial f}{\partial x_k} \frac{\partial x_k}{\partial y_k} = \frac{\partial f}{\partial x_k} \exp(y_k) = \frac{\partial f}{\partial x_k} x_k \]

Implementation Algorithm

1. mapToReal() performs element-wise exponential transformation if log-scaling is enabled for a given parameter index.
2. mapGradientToOpt() multiplies each computed gradient entry \(\frac{\partial f}{\partial x_k}\) by \(x_k\) if log-scaling is active.
3. NLoptCallback coordinates the mapping of std::vector to Eigen::Map mapping and triggers evaluations.

Constructor & Destructor Documentation

ObjectiveFunctor() [1/2]

cmp::ObjectiveFunctor::ObjectiveFunctor ( std::function< double(Eigen::Ref< const Eigen::VectorXd >)>  fval )

inlineexplicit

ObjectiveFunctor() [2/2]

cmp::ObjectiveFunctor::ObjectiveFunctor ( std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)>  fval_grad )

inlineexplicit

Member Function Documentation

addEqualityConstraint()

void cmp::ObjectiveFunctor::addEqualityConstraint ( std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)>  h )

inline

addInequalityConstraint()

void cmp::ObjectiveFunctor::addInequalityConstraint ( std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)>  g )

inline

evaluateConstraint()

double cmp::ObjectiveFunctor::evaluateConstraint ( size_t  index, bool  is_ineq, Eigen::Ref< const Eigen::VectorXd >  x_opt, Eigen::Ref< Eigen::VectorXd >  grad_opt  ) const

inlineprivate

Evaluates a constraint while properly translating inputs and mapping output gradients.

Parameters

index	Index of the constraint in the vector.
is_ineq	True if inequality, false if equality constraint.
x_opt	Input vector in optimization space.
grad_opt	Output gradient vector in optimization space.

Returns

The evaluated constraint value.

getEqualityConstraints()

std::vector< std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> > cmp::ObjectiveFunctor::getEqualityConstraints ( ) const

inline

getInequalityConstraints()

std::vector< std::function< double(Eigen::Ref< const Eigen::VectorXd >, Eigen::Ref< Eigen::VectorXd >)> > cmp::ObjectiveFunctor::getInequalityConstraints ( ) const

inline

getLogScale()

const std::vector< bool > & cmp::ObjectiveFunctor::getLogScale ( ) const

inline

mapGradientToOpt()

void cmp::ObjectiveFunctor::mapGradientToOpt ( Eigen::Ref< const Eigen::VectorXd >  x_real, Eigen::Ref< Eigen::VectorXd >  grad_real  ) const

inlineprivate

Transforms gradients from real space back to optimization space using the chain rule.

Mathematical Formulation By the chain rule, the gradient with respect to the optimization parameter is:

\[ \frac{\partial f}{\partial \theta_{\text{opt}, i}} = \frac{\partial f}{\partial \theta_{\text{real}, i}} \frac{d \theta_{\text{real}, i}}{d \theta_{\text{opt}, i}} = \frac{\partial f}{\partial \theta_{\text{real}, i}} \theta_{\text{real}, i} \]

Parameters

x_real	Parameter vector in real space.
grad_real	Gradient vector in real space (modified in-place to optimization space gradient).

mapToReal()

Eigen::VectorXd cmp::ObjectiveFunctor::mapToReal ( Eigen::Ref< const Eigen::VectorXd >  x_opt ) const

inlineprivate

Maps parameters from optimization space (potentially log-scaled) to physical space.

Mathematical Formulation For dimensions flagged as log-scaled:

\[ \theta_{\text{real}, i} = \exp\left( \theta_{\text{opt}, i} \right) \]

For unscaled dimensions:

\[ \theta_{\text{real}, i} = \theta_{\text{opt}, i} \]

Parameters

x_opt

Parameter vector in optimization space.

Returns

Parameter vector in real space.

NLoptCallback()

static double cmp::ObjectiveFunctor::NLoptCallback ( const std::vector< double > &  x, std::vector< double > &  grad, void *  data  )

inlinestatic

NLoptConstraintWrapper()

static double cmp::ObjectiveFunctor::NLoptConstraintWrapper ( const std::vector< double > &  x, std::vector< double > &  grad, void *  data  )

inlinestatic

operator()()

double cmp::ObjectiveFunctor::operator() ( Eigen::Ref< const Eigen::VectorXd >  x_opt, Eigen::Ref< Eigen::VectorXd >  grad_opt  ) const

inline

setLogScale()

void cmp::ObjectiveFunctor::setLogScale ( const std::vector< bool > &  log_scale )

inline

Set which parameters should be optimized in log-space.

Parameters

log_scale

A boolean mask where true means the parameter is log-scaled.

usesGradient()

bool cmp::ObjectiveFunctor::usesGradient ( ) const

inline

Checks if the functor utilizes gradient information.

Returns

True if gradients are used, false otherwise.

Member Data Documentation

eq_constraints_

std::vector<std::function<double(Eigen::Ref<const Eigen::VectorXd>, Eigen::Ref<Eigen::VectorXd>)> > cmp::ObjectiveFunctor::eq_constraints_

private

Collection of equality constraint functions.

fval_grad_inplace_

std::function<double(Eigen::Ref<const Eigen::VectorXd>, Eigen::Ref<Eigen::VectorXd>)> cmp::ObjectiveFunctor::fval_grad_inplace_

private

Pointer to a gradient-based objective function.

fval_only_

std::function<double(Eigen::Ref<const Eigen::VectorXd>)> cmp::ObjectiveFunctor::fval_only_

private

Pointer to a gradient-free objective function.

ineq_constraints_

std::vector<std::function<double(Eigen::Ref<const Eigen::VectorXd>, Eigen::Ref<Eigen::VectorXd>)> > cmp::ObjectiveFunctor::ineq_constraints_

private

Collection of inequality constraint functions.

log_scale_

std::vector<bool> cmp::ObjectiveFunctor::log_scale_

private

Mask indicating which parameter dimensions are optimized in log-scale.

use_gradient_

bool cmp::ObjectiveFunctor::use_gradient_

private

Flag indicating whether the objective function uses gradient information.

---

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

• include/optimization.h