#include <ConstantBandwidthSmoother1D.hh>

Inheritance diagram for npstat::ConstantBandwidthSmoother1D:

Public Member Functions
	ConstantBandwidthSmoother1D (unsigned nbins, double xmin, double xmax, int symbetaPower, unsigned kernelOrder, double bandwidth=0.0, double bwFactor=1.0, bool mirrorData=true, const char *label=0)

int	symbetaPower () const

unsigned	kernelOrder () const

double	fixedBandwidth () const

double	bwFactor () const

bool	mirrorsData () const

unsigned	dataLen () const

double	selfContribution (unsigned) const

Public Member Functions inherited from npstat::AbsMarginalSmootherBase
	AbsMarginalSmootherBase (unsigned nbins, double xmin, double xmax, const char *axisLabel=0)

void	setAxisLabel (const char *axisLabel)

unsigned	nBins () const

double	xMin () const

double	xMax () const

double	binWidth () const

const std::string &	getAxisLabel () const

gs::AbsArchive *	getArchive () const

const std::string &	getArchiveCategory () const

double	lastBandwidth () const

void	setArchive (gs::AbsArchive ar, const char category=0)

void	unsetArchive ()

template<typename Numeric >
const HistoND< double > &	smooth (const std::vector< Numeric > &inputPoints, double minValue=-std::numeric_limits< double >::max(), double maxValue=std::numeric_limits< double >::max(), double *bandwidthUsed=0)

template<typename Numeric >
const HistoND< double > &	smooth (unsigned long uniqueId, unsigned dimNumber, const std::vector< Numeric > &inputPoints, double minValue=-std::numeric_limits< double >::max(), double maxValue=std::numeric_limits< double >::max(), double *bandwidthUsed=0)

template<typename Numeric >
const HistoND< double > &	weightedSmooth (const std::vector< std::pair< Numeric, double > > &inputPoints, double minValue=-std::numeric_limits< double >::max(), double maxValue=std::numeric_limits< double >::max(), double *bandwidthUsed=0)

template<typename Numeric >
const HistoND< double > &	weightedSmooth (unsigned long uniqueId, unsigned multivariatePointDimNumber, const std::vector< std::pair< Numeric, double > > &inputPoints, double minValue=-std::numeric_limits< double >::max(), double maxValue=std::numeric_limits< double >::max(), double *bandwidthUsed=0)

Detailed Description

1-d KDE implementation with constant bandwidth, using kernels from the beta family (or the Gaussian). Based on FFT, so it should be reasonably fast even if the number of bins is large. By default, the data is mirrored at the boundary to reduce the edge effects.

The output density is truncated non-negative and normalized.

Constructor & Destructor Documentation

◆ ConstantBandwidthSmoother1D()

npstat::ConstantBandwidthSmoother1D::ConstantBandwidthSmoother1D	(	unsigned	nbins,
		double	xmin,
		double	xmax,
		int	symbetaPower,
		unsigned	kernelOrder,
		double	bandwidth = `0.0`,
		double	bwFactor = `1.0`,
		bool	mirrorData = `true`,
		const char *	label = `0`
	)

Constructor arguments are as follows:

nbins, xmin, xmax – Parameters for the histogram which will be accumulated using the data sample to be smoothed. nbins should preferrably be a power of 2 (for subsequent FFT).

symbetaPower – Power of the symmetric beta kernel to use. Gaussian kernel chopped of at 12 sigma will be used in case this parameter is negative. This parameter must not exceed 10.

kernelOrder – Order of the kernel. Meaningful arguments are even numbers 2 or larger. Numbers below 2 will be converted to 2, 1 will be subtracted from odd numbers (poly degree is the kernel order minus 2).

bandwidth – Fixed bandwidth to use. Value of 0.0 (default) means to use the Gaussian plug-in estimate which will change from one dataset to another. Note that the bin width of the histogram with which "smoothHisto" method will be called is not necessarily the same from one call to another, so the usage of fixed bandwidth has to take this into account (basically, by also using the class DummyResponseBoxBuilder in the appropriate place which will lead to constant bin width).

bwFactor – Fudge factor for the bandwidth used for the density estimate (either for the plug-in bandwidth or for the fixed one provided by the previous argument).

mirrorData – If true, the data will be mirrored at the boundary.

label – Label for the axis. Useful in case smoothing results are stored for inspection.

Member Function Documentation

◆ dataLen()

unsigned npstat::ConstantBandwidthSmoother1D::dataLen ( ) const

inlinevirtual

Method that has to be overriden from AbsPolyFilter1D

Implements npstat::AbsPolyFilter1D.

◆ selfContribution()

double npstat::ConstantBandwidthSmoother1D::selfContribution ( unsigned ) const

inlinevirtual

Method that has to be overriden from AbsPolyFilter1D. If the bandwidth is not fixed, the value returned by "selfContribution" will change after each call to "smooth" or "weightedSmooth".

Implements npstat::AbsPolyFilter1D.

◆ symbetaPower()

int npstat::ConstantBandwidthSmoother1D::symbetaPower ( ) const

inline

Simple inspector of object properties

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

stat/ConstantBandwidthSmoother1D.hh