lsst::meas::modelfit::OptimizerObjective Class Reference

Base class for objective functions for Optimizer. More...

#include <optimizer.h>

Public Member Functions
	OptimizerObjective (int dataSize_, int parameterSize_)
	Base class constructor; must be called by all subclasses.
void	fillObjectiveValueGrid (ndarray::Array< Scalar const, 2, 1 > const &parameters, ndarray::Array< Scalar, 1, 1 > const &output) const
	Evaluate the Objective on a 1-d grid.
virtual void	computeResiduals (ndarray::Array< Scalar const, 1, 1 > const &parameters, ndarray::Array< Scalar, 1, 1 > const &residuals) const =0
	Evaluate the residuals of the model for a given parameter vector.
virtual bool	differentiateResiduals (ndarray::Array< Scalar const, 1, 1 > const &parameters, ndarray::Array< Scalar, 2,-2 > const &derivatives) const
	Evaluate analytic derivatives of the model or signal that they are not available.
virtual bool	hasPrior () const
	Return true if the Objective has a Bayesian prior as well as a likelihood.
virtual Scalar	computePrior (ndarray::Array< Scalar const, 1, 1 > const &parameters) const
	Compute the value of the Bayesian prior for the given parameter vector.
virtual void	differentiatePrior (ndarray::Array< Scalar const, 1, 1 > const &parameters, ndarray::Array< Scalar, 1, 1 > const &gradient, ndarray::Array< Scalar, 2, 1 > const &hessian) const
	Compute the first and second derivatives of the Bayesian prior with respect to the parameters.
virtual	~OptimizerObjective ()

Static Public Member Functions
static std::shared_ptr< OptimizerObjective >	makeFromLikelihood (std::shared_ptr< Likelihood > likelihood, std::shared_ptr< Prior > prior=std::shared_ptr< Prior >())
	Return a concrete Objective object built from a Likelihood and Prior.

Public Attributes
int const	dataSize
int const	parameterSize

Detailed Description

Base class for objective functions for Optimizer.

Definition at line 44 of file optimizer.h.

Constructor & Destructor Documentation

OptimizerObjective()

lsst::meas::modelfit::OptimizerObjective::OptimizerObjective ( int dataSize_, int parameterSize_ )

inline

Base class constructor; must be called by all subclasses.

Definition at line 67 of file optimizer.h.

                                                          :
        dataSize(dataSize_), parameterSize(parameterSize_)
    {}

~OptimizerObjective()

virtual lsst::meas::modelfit::OptimizerObjective::~OptimizerObjective ( )

inlinevirtual

Definition at line 167 of file optimizer.h.

{}

Member Function Documentation

computePrior()

virtual Scalar lsst::meas::modelfit::OptimizerObjective::computePrior ( ndarray::Array< Scalar const, 1, 1 > const & parameters ) const

inlinevirtual

Compute the value of the Bayesian prior for the given parameter vector.

The default implementation simply returns 1.0 (appropriate for an unnormalized constant prior, which is mathematically equivalent to no prior).

Definition at line 142 of file optimizer.h.

{ return 1.0; }

computeResiduals()

virtual void lsst::meas::modelfit::OptimizerObjective::computeResiduals ( ndarray::Array< Scalar const, 1, 1 > const & parameters, ndarray::Array< Scalar, 1, 1 > const & residuals ) const

pure virtual

Evaluate the residuals of the model for a given parameter vector.

Parameters

[in]	parameters	An array of parameters with shape (parameterSize).
[out]	residuals	Output array that will contain (model - data) on return. Must be allocated to shape (dataSize), but need not be initialized.

differentiatePrior()

virtual void lsst::meas::modelfit::OptimizerObjective::differentiatePrior ( ndarray::Array< Scalar const, 1, 1 > const & parameters, ndarray::Array< Scalar, 1, 1 > const & gradient, ndarray::Array< Scalar, 2, 1 > const & hessian ) const

inlinevirtual

Compute the first and second derivatives of the Bayesian prior with respect to the parameters.

The default implementation simply sets the output arrays to 0.0 (appropriate for an constant prior, which is mathematically equivalent to no prior).

Parameters

[in]	parameters	An array of parameters with shape (parameterSize).
[out]	gradient	First derivative of the prior with respect to the parameters. Must be allocated to shape (parameterSize), but need not be initialized.
[out]	hessian	Second derivative of the prior with respect to the parameters. Must be allocated to shape (parameterSize, parameterSize), but need not be initialized. This is a symmetric matrix, and only the lower triangle is used.

Definition at line 158 of file optimizer.h.

            {
        gradient.deep() = 0.0;
        hessian.deep() = 0.0;
    }

differentiateResiduals()

virtual bool lsst::meas::modelfit::OptimizerObjective::differentiateResiduals ( ndarray::Array< Scalar const, 1, 1 > const & parameters, ndarray::Array< Scalar, 2,-2 > const & derivatives ) const

inlinevirtual

Evaluate analytic derivatives of the model or signal that they are not available.

Subclasses that provide analytic derivatives should implement this method and return true. Subclasses that do not should return false to indicate to the optimizer that numeric derivatives must be used instead. Subclasses that can only sometimes compute analytic derivatives are not supported.

Parameters

[in]	parameters	An array of parameters with shape (parameterSize).
[out]	derivatives	Output array that will contain d(model - data)/d(parameters) on return. Must be allocated to shape (dataSize, parameterSize), but need not be initialized.

Definition at line 121 of file optimizer.h.

            {
        return false;
    }

fillObjectiveValueGrid()

void lsst::meas::modelfit::OptimizerObjective::fillObjectiveValueGrid	(	ndarray::Array< Scalar const, 2, 1 > const &	parameters,
		ndarray::Array< Scalar, 1, 1 > const &	output ) const

Evaluate the Objective on a 1-d grid.

This delegates to computeResiduals, and hence need not be reimplemented by derived classes. It is intended mostly for diagnostic purposes; when investigating the behavior of the optimizer, it is frequently valuable to plot its path against the gridded objective function.

Parameters

[in]	parameters	An array with shape (gridSize, parameterSize), specifying the parameter vector at points on the grid.
[out]	output	Output array for objective values with shape (gridSize). Must be allocated, but need not be initialized.

Frequently, the arguments to this function will be flattened views into higher dimensional arrays, allowing it to be used to construct N-d grids despite taking only a 1-d sequence of grid points.

hasPrior()

virtual bool lsst::meas::modelfit::OptimizerObjective::hasPrior ( ) const

inlinevirtual

Return true if the Objective has a Bayesian prior as well as a likelihood.

The default implementation returns false.

Definition at line 134 of file optimizer.h.

{ return false; }

makeFromLikelihood()

static std::shared_ptr< OptimizerObjective > lsst::meas::modelfit::OptimizerObjective::makeFromLikelihood ( std::shared_ptr< Likelihood > likelihood, std::shared_ptr< Prior > prior = std::shared_ptr< Prior >() )

static

Return a concrete Objective object built from a Likelihood and Prior.

Most fitting problems that can be formulated in terms of (multi-shapelet) Models, Likelihoods, and Priors can just use this Objective. The returned Objective relies on numerical derivatives, however, so simple problems where analytic derivatives are easy to implement may merit a custom OptimizerObjective.

Member Data Documentation

dataSize

int const lsst::meas::modelfit::OptimizerObjective::dataSize

Definition at line 47 of file optimizer.h.

parameterSize

int const lsst::meas::modelfit::OptimizerObjective::parameterSize

Definition at line 48 of file optimizer.h.

---

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

• /j/snowflake/release/lsstsw/stack/lsst-scipipe-9.0.0/Linux64/meas_modelfit/ga5288a1d22+2a84bb7594/include/lsst/meas/modelfit/optimizer.h