ai.libs.jaicore.ml.tsc.classifier.shapelets

Class LearnShapeletsLearningAlgorithm

java.lang.Object

ai.libs.jaicore.basic.algorithm.AAlgorithm<TimeSeriesDataset,C>

ai.libs.jaicore.ml.tsc.classifier.ASimplifiedTSCLearningAlgorithm<java.lang.Integer,LearnShapeletsClassifier>

ai.libs.jaicore.ml.tsc.classifier.shapelets.LearnShapeletsLearningAlgorithm

All Implemented Interfaces:

ai.libs.jaicore.basic.algorithm.IAlgorithm<TimeSeriesDataset,LearnShapeletsClassifier>, ai.libs.jaicore.basic.Cancelable, ai.libs.jaicore.basic.ILoggingCustomizable, java.lang.Iterable<ai.libs.jaicore.basic.algorithm.events.AlgorithmEvent>, java.util.concurrent.Callable<LearnShapeletsClassifier>, java.util.Iterator<ai.libs.jaicore.basic.algorithm.events.AlgorithmEvent>

public class LearnShapeletsLearningAlgorithm
extends ASimplifiedTSCLearningAlgorithm<java.lang.Integer,LearnShapeletsClassifier>

Generalized Shapelets Learning implementation for LearnShapeletsClassifier published in "J. Grabocka, N. Schilling, M. Wistuba, L. Schmidt-Thieme: Learning Time-Series Shapelets" (https://www.ismll.uni-hildesheim.de/pub/pdfs/grabocka2014e-kdd.pdf)

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static interface	LearnShapeletsLearningAlgorithm.ILearnShapeletsLearningAlgorithmConfig

Field Summary

Fields

Modifier and Type	Field and Description
static double	ALPHA Predefined alpha parameter used within the calculations.
static boolean	USE_BIAS_CORRECTION Indicator whether Bessel's correction should be used when normalizing arrays.

Constructor Summary

Constructors

Constructor and Description
LearnShapeletsLearningAlgorithm(LearnShapeletsLearningAlgorithm.ILearnShapeletsLearningAlgorithmConfig config, LearnShapeletsClassifier classifier, TimeSeriesDataset dataset) Constructor of the algorithm to train a LearnShapeletsClassifier.

Method Summary

Modifier and Type	Method and Description
static double	calculateD(double[][][] s, int minShapeLength, int r, double[] instance, int k, int j) Function to calculate the distance between the j-th segment of the given time series instance and the k-th shapelet stored in the shapelet tensor S.
static double	calculateMHat(double[][][] s, int minShapeLength, int r, double[] instance, int k, int Q, double alpha) Function to calculate the soft-minimum function which is a differentiable approximation of the minimum distance matrix given in the paper in section 3.1.4.
LearnShapeletsClassifier	call() Main function to train a LearnShapeletsClassifier.
int	getC()
LearnShapeletsLearningAlgorithm.ILearnShapeletsLearningAlgorithmConfig	getConfig()
static int	getNumberOfSegments(int Q, int minShapeLength, int r) Returns the number of segments which are available for a instance with Q attributes for a given scale r and a minimum shape length minShapeLength.
double[][][]	initializeS(double[][] trainingMatrix) Initializes the tensor S storing the shapelets for each scale.
void	initializeWeights(double[][][] w, double[] w0) Randomly initializes the weights around zero.
boolean	isUseInstanceReordering()
ai.libs.jaicore.basic.algorithm.events.AlgorithmEvent	nextWithException()
void	performSGD(double[][][] w, double[][][] wHist, double[] w0, double[] w0Hist, double[][][] s, double[][][] sHist, double[][] dataMatrix, int[][] y, long beginTime, int[] targets) Method performing the stochastic gradient descent to learn the weights and shapelets.
void	setC(int c)
void	setUseInstanceReordering(boolean useInstanceReordering)
java.util.List<java.lang.Integer>	shuffleAccordingToAlternatingClassScheme(java.util.List<java.lang.Integer> instanceIndices, int[] targets, java.util.Random random) Shuffles the data in a class alternating scheme.

Methods inherited from class ai.libs.jaicore.ml.tsc.classifier.ASimplifiedTSCLearningAlgorithm

cancel, getClassifier, hasNext, iterator, next, registerListener

Methods inherited from class ai.libs.jaicore.basic.algorithm.AAlgorithm

activate, announceTimeoutDetected, avoidReinterruptionOnShutdownOnCurrentThread, checkAndConductTermination, checkTermination, computeTimeoutAware, getActivationTime, getId, getInput, getLoggerName, getNumCPUs, getRemainingTimeToDeadline, getState, getTimeout, getTimeoutPrecautionOffset, hasThreadBeenInterruptedDuringShutdown, interruptThreadAsPartOfShutdown, isCanceled, isShutdownInitialized, isStopCriterionSatisfied, isTimeouted, post, registerActiveThread, resolveShutdownInterruptOnCurrentThread, setConfig, setLoggerName, setMaxNumThreads, setNumCPUs, setState, setTimeout, setTimeout, setTimeoutPrecautionOffset, shutdown, terminate, unregisterActiveThread, unregisterThreadAndShutdown

Methods inherited from class java.lang.Object

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

Methods inherited from interface java.lang.Iterable

forEach, spliterator

Methods inherited from interface java.util.Iterator

forEachRemaining, remove

Field Detail

USE_BIAS_CORRECTION

public static final boolean USE_BIAS_CORRECTION

Indicator whether Bessel's correction should be used when normalizing arrays.

See Also:

Constant Field Values

ALPHA

public static final double ALPHA

Predefined alpha parameter used within the calculations.

See Also:

Constant Field Values

Constructor Detail

LearnShapeletsLearningAlgorithm

public LearnShapeletsLearningAlgorithm(LearnShapeletsLearningAlgorithm.ILearnShapeletsLearningAlgorithmConfig config,
                                       LearnShapeletsClassifier classifier,
                                       TimeSeriesDataset dataset)

Constructor of the algorithm to train a LearnShapeletsClassifier.

Method Detail

initializeS

public double[][][] initializeS(double[][] trainingMatrix)
                         throws TrainingException

Initializes the tensor S storing the shapelets for each scale. The initialization is done by deriving inital shapelets from all normalized segments.

Parameters:

trainingMatrix - The training matrix used for the initialization of S.

Returns:

Return the initialized tensor storing an initial guess for the shapelets based on the clustering

Throws:

TrainingException

call

public LearnShapeletsClassifier call()
                              throws ai.libs.jaicore.basic.algorithm.exceptions.AlgorithmException

Main function to train a LearnShapeletsClassifier.

Throws:

ai.libs.jaicore.basic.algorithm.exceptions.AlgorithmException

initializeWeights

public void initializeWeights(double[][][] w,
                              double[] w0)

Randomly initializes the weights around zero. As opposed to the paper, the approach has been changed to a different standard deviation as used in the reference implementation for performance reasons.

Parameters:

w - The weight matrix

w0 - The bias vector

performSGD

public void performSGD(double[][][] w,
                       double[][][] wHist,
                       double[] w0,
                       double[] w0Hist,
                       double[][][] s,
                       double[][][] sHist,
                       double[][] dataMatrix,
                       int[][] y,
                       long beginTime,
                       int[] targets)

Method performing the stochastic gradient descent to learn the weights and shapelets.

Parameters:

w - The weight matrix

wHist - The weight's history matrix used for smoothing learning

w0 - The bias vector

w0Hist - The bias' history vector used for smoothing learning

s - The shapelet matrix

sHist - The shapelet's history matrix used for smoothing learning

dataMatrix - The data values matrix

y - The binarized target matrix

beginTime - The begin time used to check for the timeout

shuffleAccordingToAlternatingClassScheme

public java.util.List<java.lang.Integer> shuffleAccordingToAlternatingClassScheme(java.util.List<java.lang.Integer> instanceIndices,
                                                                                  int[] targets,
                                                                                  java.util.Random random)

Shuffles the data in a class alternating scheme. That means that at first, all indices per class are shuffled. Then, the randomized indices are selected in a round robin fashion among the classes.

Parameters:

instanceIndices - The instance indices the original dataset

targets - The targets of each instance

random - Random object used for randomized shuffling

Returns:

Returns the list of the shuffled indices of the alternating class scheme. Each index of instanceIndices is only used once (without replacement)

calculateMHat

public static double calculateMHat(double[][][] s,
                                   int minShapeLength,
                                   int r,
                                   double[] instance,
                                   int k,
                                   int Q,
                                   double alpha)

Function to calculate the soft-minimum function which is a differentiable approximation of the minimum distance matrix given in the paper in section 3.1.4.

Parameters:

s - The tensor storing the shapelets for different scales

minShapeLength - The minimum shape length

r - The number of scale to look at

instance - The instance time series vector

k - The index of the shapelet to look at

Q - The number of attributes (time series length)

alpha - Parameter to control the desired precision of the M_hat approximation

Returns:

Returns the approximation of the minimum distance of the instance and the shapelet given by the parameters r and k.

calculateD

public static double calculateD(double[][][] s,
                                int minShapeLength,
                                int r,
                                double[] instance,
                                int k,
                                int j)

Function to calculate the distance between the j-th segment of the given time series instance and the k-th shapelet stored in the shapelet tensor S.

Parameters:

s - The tensor storing the shapelets for different scales

minShapeLength - The minimum shape length

r - The number of scale to look at

instance - The instance time series vector

k - The index of the shapelet to look at

j - The segment of the instance time series to look at

Returns:

Returns the minimum distance of the j-th segment of the instance and the shapelet given by the parameters r, k and j.

nextWithException

public ai.libs.jaicore.basic.algorithm.events.AlgorithmEvent nextWithException()

Specified by:

nextWithException in interface ai.libs.jaicore.basic.algorithm.IAlgorithm<TimeSeriesDataset,LearnShapeletsClassifier>

Overrides:

nextWithException in class ASimplifiedTSCLearningAlgorithm<java.lang.Integer,LearnShapeletsClassifier>

getNumberOfSegments

public static int getNumberOfSegments(int Q,
                                      int minShapeLength,
                                      int r)

Returns the number of segments which are available for a instance with Q attributes for a given scale r and a minimum shape length minShapeLength.

Parameters:

Q - Number of attributes of an instance

minShapeLength - Minimum shapelet length

r - Scale to be looked at

Returns:

Returns the number of segments which can be looked at for an instance with Q time series attributes

getConfig

public LearnShapeletsLearningAlgorithm.ILearnShapeletsLearningAlgorithmConfig getConfig()

Specified by:

getConfig in interface ai.libs.jaicore.basic.algorithm.IAlgorithm<TimeSeriesDataset,LearnShapeletsClassifier>

Overrides:

getConfig in class ai.libs.jaicore.basic.algorithm.AAlgorithm<TimeSeriesDataset,LearnShapeletsClassifier>

isUseInstanceReordering

public boolean isUseInstanceReordering()

Returns:

the useInstanceReordering

setUseInstanceReordering

public void setUseInstanceReordering(boolean useInstanceReordering)

Parameters:

useInstanceReordering - the useInstanceReordering to set

getC

public int getC()

Returns:

the c

setC

public void setC(int c)

Parameters:

c - the c to set