uk.ac.starlink.ttools.plot2.layer

Interface Kernel1dShape

All Known Implementing Classes:

StandardKernel1dShape

public interface Kernel1dShape

Factory interface for for Kernel1d smoothing functional forms.

Some implementations are provided in the StandardKernel1dShape class.

Since:

2 Mar 2015

Author:

Mark Taylor

Method Summary

Modifier and Type	Method and Description
Kernel1d	createFixedWidthKernel(double width) Creates a fixed width kernel with a given nominal width.
Kernel1d	createKnnKernel(double k, boolean isSymmetric, int minWidth, int maxWidth) Creates an adaptive kernel that uses a K-nearest-neighbours algorithm to determine local smoothing width, so that the width of the kernel is determined by the distance (number of 1-pixel bins) within which the given number k of samples is found.
Kernel1d	createMeanKernel(double width) Creates an averaging kernel with a given nominal fixed width.
String	getDescription() Returns a short description for this shape.
String	getName() Returns a one-word name for this shape.

Method Detail

getName

String getName()

Returns a one-word name for this shape.

Returns:

name

getDescription

String getDescription()

Returns a short description for this shape.

Returns:

description

createFixedWidthKernel

Kernel1d createFixedWidthKernel(double width)

Creates a fixed width kernel with a given nominal width. The width is some kind of characteristic half-width in one direction of the smoothing function. It is in units of grid points (array element spacing). It would generally be less than or equal to the kernel's extent.

Parameters:

width - half-width

Returns:

new kernel

createMeanKernel

Kernel1d createMeanKernel(double width)

Creates an averaging kernel with a given nominal fixed width. The 'convolution' it performs is not really a convolution, instead it's a sort of weighted moving average. This is a smoothing that's suitable for intensive quantities. Using proper convolution for intensive quantities like the mean or median is problematic if there may be blank values in the input array, since the smoothed value has to keep track of how many non-blank values it has encountered.

Parameters:

width - half-width

Returns:

new kernel

createKnnKernel

Kernel1d createKnnKernel(double k,
                         boolean isSymmetric,
                         int minWidth,
                         int maxWidth)

Creates an adaptive kernel that uses a K-nearest-neighbours algorithm to determine local smoothing width, so that the width of the kernel is determined by the distance (number of 1-pixel bins) within which the given number k of samples is found.

The nearest neighbour search may be symmetric or asymmetric. In the asymmetric case, the kernel width is determined separately for the positive and negative directions along the axis.

Minimum and maximum smoothing widths are also supplied as bounds on the smoothing width for the case that the samples are very dense or very spread out (the latter case covers the edge of the data region as well). If minWidth==maxWidth, the result is a fixed-width kernel.

Parameters:

k - number of nearest neighbours included in the distance that characterises the smoothing

isSymmetric - true for bidirectional KNN search, false for unidirectional

minWidth - minimum smoothing width

maxWidth - maximum smoothing width

Returns:

new kernel