ai.libs.jaicore.ml.tsc.distances

Class TransformDistance

java.lang.Object

ai.libs.jaicore.ml.tsc.distances.TransformDistance

All Implemented Interfaces:

ITimeSeriesDistance

public class TransformDistance
extends java.lang.Object
implements ITimeSeriesDistance

Implementation of the Transform Distance (TD) measure as published in "Non-isometric transforms in time series classification using DTW" by Tomasz Gorecki and Maciej Luczak (2014). Building up on their work of the Derivate Distance, the authors were looking for functions other than the derivative which can be used in a similar manner, what leads them to use transform instead of derivates. Given a distance measure d, the Transform Distance for the two time series A and B is: a * d(A, B) + b * d(t(A), t(B)), where t(A) and t(B) are transforms (@see jaicore.ml.tsc.filter.transform) of A and B respec. and 0 <= a <= 1, 0 <= b <= 1> are parameters of the measure. The parameters a and b are set via an alpha value, that is a=cos(alpha) and b=sin(alpha). The Transform Distance that uses Dynamic Time Warping as underlying distance measure is commonly denoted as TD_DTW. The Transform Distance that uses the Euclidean distance as underlying distance measure is commonly denoted as TD_ED.

It is also possible to use a distinct distance measure to calculate the distance between the time series and its transforms.

Constructor Summary

Constructors

Constructor and Description
TransformDistance(double alpha, ATransformFilter transform, ITimeSeriesDistance distance) Constructor that uses the same distance measures for the function and transform values.
TransformDistance(double alpha, ATransformFilter transform, ITimeSeriesDistance timeSeriesDistance, ITimeSeriesDistance transformDistance) Constructor with individual distance measures for the function and transform values.
TransformDistance(double alpha, ITimeSeriesDistance distance) Constructor that uses the same distance measures for the function and transform values that uses the CosineTransform as transformation.
TransformDistance(double alpha, ITimeSeriesDistance timeSeriesDistance, ITimeSeriesDistance transformDistance) Constructor with individual distance measures for the function and transform values that uses the CosineTransform as transformation.

Method Summary

Modifier and Type	Method and Description
double	distance(double[] a, double[] b) Calculates the distance between two time series.
double	getA() Getter for the a parameter.
double	getAlpha() Getter for the alpha value.
double	getB() Getter for the a parameter.
void	setAlpha(double alpha) Sets the alpha value and adjusts the measurement parameters a = cos(alpha) and b = sin(alpha) accordingly.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TransformDistance

public TransformDistance(double alpha,
                         ATransformFilter transform,
                         ITimeSeriesDistance timeSeriesDistance,
                         ITimeSeriesDistance transformDistance)

Constructor with individual distance measures for the function and transform values.

Parameters:

alpha - @see #alpha ,0 <= alpha <= pi/2.

transform - The transform calculation to use.

timeSeriesDistance - The distance measure to use to calculate the distance of the transform values.

transformDistance - The distance measure to use to calculate the distance of the transform values.

TransformDistance

public TransformDistance(double alpha,
                         ITimeSeriesDistance timeSeriesDistance,
                         ITimeSeriesDistance transformDistance)

Constructor with individual distance measures for the function and transform values that uses the CosineTransform as transformation.

Parameters:

alpha - The distance measure to use to calculate the distance of the function values. 0 <= alpha <= pi/2.

timeSeriesDistance - The distance measure to use to calculate the distance of the transform values.

transformDistance - The distance measure to use to calculate the distance of the transform values.

TransformDistance

public TransformDistance(double alpha,
                         ATransformFilter transform,
                         ITimeSeriesDistance distance)

Constructor that uses the same distance measures for the function and transform values.

Parameters:

alpha - The distance measure to use to calculate the distance of the function values. 0 <= alpha <= pi/2.

transform - The transform calculation to use.

distance - The distance measure to use to calculate the distance of the function and transform values.

TransformDistance

public TransformDistance(double alpha,
                         ITimeSeriesDistance distance)

Constructor that uses the same distance measures for the function and transform values that uses the CosineTransform as transformation.

Parameters:

alpha - The distance measure to use to calculate the distance of the function values. 0 <= alpha <= pi/2.

distance - The distance measure to use to calculate the distance of the function and transform values.

Method Detail

distance

public double distance(double[] a,
                       double[] b)

Description copied from interface: ITimeSeriesDistance

Calculates the distance between two time series.

Specified by:

distance in interface ITimeSeriesDistance

Parameters:

a - First time series.

b - Second time series.

Returns:

Distance between the first and second time series.

setAlpha

public void setAlpha(double alpha)

Sets the alpha value and adjusts the measurement parameters a = cos(alpha) and b = sin(alpha) accordingly.

Parameters:

alpha - The alpha value, 0 <= alpha <= pi/2.

getAlpha

public double getAlpha()

Getter for the alpha value. It holds, that 0 <= alpha <= pi/2.

getA

public double getA()

Getter for the a parameter. @see #a

getB

public double getB()

Getter for the a parameter. @see #b