#include <LocalQuadraticLeastSquaresND.hh>

Public Member Functions
	LocalQuadraticLeastSquaresND (unsigned polyDegree, const AbsDistributionND &weight, const double bandwidthValues, unsigned nBandwidthValues, const Point pointCoords, unsigned nPoints, const Numeric values, const Numeric errors=0)

unsigned	dim () const

unsigned	polyDegree () const

unsigned	nPoints () const

unsigned	nTermsFitted () const

bool	hasErrors () const

double	getBandwidth (const unsigned bwNumber) const

double	chiSquare () const

double	lscvCriterion () const

void	setPointCoords (const Point *newCoords)

void	setValues (const Numeric *newValues)

void	setErrors (const Numeric *newErrors)

void	setBandwidth (const unsigned bwNum, const double newBw)

Matrix< double >	smootherMatrix (double stepInErrorUnits=1.0e-3) const

template<typename Num2 >
double	fit (const Num2 coordinate, unsigned coordinateDim, double gradient=0, unsigned lenGradient=0, double *hessian=0, unsigned lenHessian=0, unsigned excludedPoint=UINT_MAX) const

Detailed Description

template<class Point, typename Numeric>
class npstat::LocalQuadraticLeastSquaresND< Point, Numeric >

Local quadratic least squares fitter. Template parameter class Point must be subscriptable.

Constructor & Destructor Documentation

◆ LocalQuadraticLeastSquaresND()

template<class Point , typename Numeric >

npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::LocalQuadraticLeastSquaresND	(	unsigned	polyDegree,
		const AbsDistributionND &	weight,
		const double *	bandwidthValues,
		unsigned	nBandwidthValues,
		const Point *	pointCoords,
		unsigned	nPoints,
		const Numeric *	values,
		const Numeric *	errors = `0`
	)

The "polyDegree" parameter must be 0, 1, or 2.

All values in the "bandwidthValues" array must be positive and their number, "nBandwidthValues", must be equal to the dimensionality of the weight function, weight.dim(). The bandwidth values will be copied into an internal buffer.

Make sure that the bandwidth values are sufficiently large so that the number of points for which the weight function is positive is always at least as large as the number of polynomial terms fitted. This must be true for every future "fit" call.

This object will not copy or own "pointCoords", "values", or "errors" arrays. The number of elements in all of these arrays must be the same and equal to "nPoints". All errors must be positive. The "errors" array can also be NULL in which case all errors are assumed to be equal.

Member Function Documentation

◆ chiSquare()

template<class Point , typename Numeric >

double npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::chiSquare ( ) const

The following function returns the chi-square of the fit. If the errors have not been specified, the sum of squared differences is returned (as if each error is set to 1).

◆ dim()

template<class Point , typename Numeric >

unsigned npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::dim ( ) const

inline

Examine object properties

◆ fit()

template<class Point , typename Numeric >

template<typename Num2 >

double npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::fit	(	const Num2 *	coordinate,
		unsigned	coordinateDim,
		double *	gradient = `0`,
		unsigned	lenGradient = `0`,
		double *	hessian = `0`,
		unsigned	lenHessian = `0`,
		unsigned	excludedPoint = `UINT_MAX`
	)		const

Perfom local quadratic polynomial fit with weight function centered at "coordinate". In additional to the fitted value, one can get the gradient and the Hessian at the fitted point, as estimated by the fit.

◆ lscvCriterion()

template<class Point , typename Numeric >

double npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::lscvCriterion ( ) const

The following function returns the chi-square obtained by the leave-one-out least squares cross validation

◆ setBandwidth()

template<class Point , typename Numeric >

void npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::setBandwidth	(	const unsigned	bwNum,
		const double	newBw
	)

inline

Change bandwidth used for weight calculations

◆ setErrors()

template<class Point , typename Numeric >

void npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::setErrors ( const Numeric * newErrors )

inline

Change errors for the point locations provided in the constructor. The array must have at least "nPoints()" elements (or it can be NULL).

◆ setPointCoords()

template<class Point , typename Numeric >

void npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::setPointCoords ( const Point * newCoords )

inline

Change the point coordinates. Note that the number of new coordinates must be exactly equal to the number of points provided in the constructor. The array will not be copied.

◆ setValues()

template<class Point , typename Numeric >

void npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::setValues ( const Numeric * newValues )

inline

Change values for the point locations provided in the constructor. The array must have at least "nPoints()" elements.

◆ smootherMatrix()

template<class Point , typename Numeric >

Matrix<double> npstat::LocalQuadraticLeastSquaresND< Point, Numeric >::smootherMatrix ( double stepInErrorUnits = 1.0e-3 ) const

Estimate of the smoother matrix. Will be nPoints x nPoints in size, so this works only for relatively small datasets. Smoother matrix plays similar role to the hat matrix.

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

stat/LocalQuadraticLeastSquaresND.hh

Public Member Functions

Detailed Description

template<class Point, typename Numeric> class npstat::LocalQuadraticLeastSquaresND< Point, Numeric >

Constructor & Destructor Documentation

◆ LocalQuadraticLeastSquaresND()

Member Function Documentation

◆ chiSquare()

◆ dim()

◆ fit()

◆ lscvCriterion()

◆ setBandwidth()

◆ setErrors()

◆ setPointCoords()

◆ setValues()

◆ smootherMatrix()

template<class Point, typename Numeric>
class npstat::LocalQuadraticLeastSquaresND< Point, Numeric >