be.ac.ulb.mlg.utils

Class DefaultRenormalizer

java.lang.Object

be.ac.ulb.mlg.utils.DefaultRenormalizer

All Implemented Interfaces:

Renormalizer<double[][]>

Direct Known Subclasses:

TaxonRenormalizer

public class DefaultRenormalizer
extends java.lang.Object
implements Renormalizer<double[][]>

DefaultRenormalizer supports custom or columns summing renormalization with features (row to be ignorated) and can handle non-overlapping groups (division of input vectors into subsets where elements of each subset are normalized only with others from the same group)

Version:

1.2, 18/08/2015

Author:

Jean-Sébastien Lerat

Field Summary

Fields

Modifier and Type	Field and Description
static double	SUM_NORMILIZER Sentinel value meaning that the renormalizer use the sum function on columns

Constructor Summary

Constructors

Constructor and Description
DefaultRenormalizer() Construct a new DefaultRenormalizer without groups, without features and with sum normalization shuffles are disable

Method Summary

Modifier and Type	Method and Description
void	addFeature(int dataIndex)
double[][]	cloneState() Avoid concurrent modification and inconsistent results when multi-threading is enabled.
int	countFeatures() Count the number of features
int	countGroups() Count the number of groups
int	groupOf(int row) Get the group of a row
boolean	hasNativeImplementation() Know if the current renormalizer owns a native implementation
boolean	isUsingCustomNormalizer() Know if the configuration of the DefaultRenormalizer uses custom renormalization values
boolean	isUsingFeatures() Know if the configuration of the DefaultRenormalizer use the feature renormalization
boolean	isUsingGroups() Know if the configuration of the DefaultRenormalizer use the group renormalization
protected void	mergeRenormalizer(double[][] sum)
void	noFeatures() Specifies to no longer use the features renormalization
void	noGroups() Specifies to no longer use the group renormalization but preserve renormalization values by group to rester them, use resetNormalizer()
double[][]	normalizeInput(double[][] input) Apply the normalization process on the given vector of data vectors
double[][]	normalizeOutput(double[][] input, double[][] output, Measure measure) Apply the normalization process on the given vector of data vectors and result measures
protected void	prepareRenormalization(double[][] sum, double[] vect, int row, boolean ignore)
boolean	processInput() Know if the renormalizer can handle input (call normalizeInput has effect)
boolean	processOutput() Know if the renormalizer can handle output (call normalizeOutput has effect)
void	resetNormalizer() Reset the custom renormalizer to use the default method (sum by columns)
void	setGroup(int row, int group) Define the group for one input data vector
void	setGroup(int row, java.lang.String groupName) Define the group for one input data vector
void	setGroups(int[] groups, int groupCount) Define the group of all input data vectors by their indexes
void	setGroupsNames(int rows, java.lang.String[] names) Define groups and associate them a label.
void	setNormalizer(int column, double divider) Define a specific value to use instead of the sum by column
void	setNormalizer(int group, int column, double divider) Define a specific value to use instead of the sum by column for groups
void	setNormalizer(java.lang.String groupName, int column, double divider) Define a specific value to use instead of the sum by column for groups
boolean	shufflePair() Know if the measurer use pre-shuffle
void	simulateRenormalization(double[][] sum, double[] shuffledFirst, double[] shuffledSecond, int first, int second) Method called when pairwise renormalization is used in order to obtain a normalized version of the shuffled vector as if it was in the original input matrix
void	updateRenormalization(double[][] sum, double[] vect, int row, boolean ignore) Method called when pairwise renormalization is used in order to ignore or not a row from the original input matrix
void	usePairShuffle(boolean shuffle) Enable or disable pairwise shuffle

Methods inherited from class java.lang.Object

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

Field Detail

SUM_NORMILIZER

public static final double SUM_NORMILIZER

Sentinel value meaning that the renormalizer use the sum function on columns

See Also:

Constant Field Values

Constructor Detail

DefaultRenormalizer

public DefaultRenormalizer()

Construct a new DefaultRenormalizer without groups, without features and with sum normalization shuffles are disable

Method Detail

addFeature

public void addFeature(int dataIndex)

isUsingGroups

public boolean isUsingGroups()

Know if the configuration of the DefaultRenormalizer use the group renormalization

Returns:

true if the DefaultRenormalizer use the group renormalization

noGroups

public void noGroups()

Specifies to no longer use the group renormalization but preserve renormalization values by group to rester them, use resetNormalizer()

resetNormalizer

public void resetNormalizer()

Reset the custom renormalizer to use the default method (sum by columns)

isUsingFeatures

public boolean isUsingFeatures()

Know if the configuration of the DefaultRenormalizer use the feature renormalization

Returns:

true if the DefaultRenormalizer use the feature renormalization

isUsingCustomNormalizer

public boolean isUsingCustomNormalizer()

Know if the configuration of the DefaultRenormalizer uses custom renormalization values

Returns:

true if the DefaultRenormalizer uses custom renormalization values

noFeatures

public void noFeatures()

Specifies to no longer use the features renormalization

setGroup

public void setGroup(int row,
                     java.lang.String groupName)

Define the group for one input data vector

Parameters:

row - The index of the input data vector (used with Measurer)

groupName - The name of the group of the input data vector

setGroup

public void setGroup(int row,
                     int group)

Define the group for one input data vector

Parameters:

row - The index of the input data vector (used with Measurer)

group - The index of the group of the input data vector

setGroupsNames

public void setGroupsNames(int rows,
                           java.lang.String[] names)

Define groups and associate them a label.

Parameters:

rows - The number of input data vectors to be processed

names - The name of all existing groups

setGroups

public void setGroups(int[] groups,
                      int groupCount)

Define the group of all input data vectors by their indexes

Parameters:

groups - The array mapping each for index i, the input data vector i to the group groups[i]

groupCount - The number of groups

setNormalizer

public void setNormalizer(java.lang.String groupName,
                          int column,
                          double divider)

Define a specific value to use instead of the sum by column for groups

Parameters:

groupName - The name of the group

column - The column index (for an input data vector v, it's i of v[i])

divider - The specific value to use as normalizer

setNormalizer

public void setNormalizer(int group,
                          int column,
                          double divider)

Define a specific value to use instead of the sum by column for groups

Parameters:

group - The index of the group

column - The column index (for an input data vector v, it's i of v[i])

divider - The specific value to use as normalizer

setNormalizer

public void setNormalizer(int column,
                          double divider)

Define a specific value to use instead of the sum by column

Parameters:

column - The column index (for an input data vector v, it's i of v[i])

divider - The specific value to use as normalizer

countFeatures

public int countFeatures()

Count the number of features

Returns:

The number of features

countGroups

public int countGroups()

Count the number of groups

Returns:

The number of groups

groupOf

public int groupOf(int row)

Get the group of a row

Parameters:

row - The row index

Returns:

The group identifier

mergeRenormalizer

protected void mergeRenormalizer(double[][] sum)

normalizeInput

public double[][] normalizeInput(double[][] input)

Description copied from interface: Renormalizer

Apply the normalization process on the given vector of data vectors

Specified by:

normalizeInput in interface Renormalizer<double[][]>

Parameters:

input - The given vector of data vectors

Returns:

the input array

updateRenormalization

public void updateRenormalization(double[][] sum,
                                  double[] vect,
                                  int row,
                                  boolean ignore)

Description copied from interface: Renormalizer

Method called when pairwise renormalization is used in order to ignore or not a row from the original input matrix

Specified by:

updateRenormalization in interface Renormalizer<double[][]>

Parameters:

sum - The state on which modifications will be applied

vect - A vector which is considerate as a row from the original input matrix corresponding to the row at the given index

row - The index of the vector to map it into the original matrix at this index

ignore - A boolean to tell if the values of the vector must be ignored (typically remove these values from the normalization process as if it was the row "index" from the original input matrix) with the true value (typically false to add it into renormalization)

prepareRenormalization

protected void prepareRenormalization(double[][] sum,
                                      double[] vect,
                                      int row,
                                      boolean ignore)

simulateRenormalization

public void simulateRenormalization(double[][] sum,
                                    double[] shuffledFirst,
                                    double[] shuffledSecond,
                                    int first,
                                    int second)

Description copied from interface: Renormalizer

Method called when pairwise renormalization is used in order to obtain a normalized version of the shuffled vector as if it was in the original input matrix

Specified by:

simulateRenormalization in interface Renormalizer<double[][]>

Parameters:

sum - The state on which modifications will be applied

shuffledFirst - The first vector of the pairwise comparison (shuffled version)

shuffledSecond - The second vector of the pairwise comparison (shuffled version)

first - The index of the first vector from the original input matrix

second - The index of the second vector from the original input matrix

normalizeOutput

public double[][] normalizeOutput(double[][] input,
                                  double[][] output,
                                  Measure measure)

Description copied from interface: Renormalizer

Apply the normalization process on the given vector of data vectors and result measures

Specified by:

normalizeOutput in interface Renormalizer<double[][]>

Parameters:

input - The given vector of data vectors

output - The result vector (measures of data vectors)

measure - The corresponding measure

Returns:

the output array

hasNativeImplementation

public boolean hasNativeImplementation()

Description copied from interface: Renormalizer

Know if the current renormalizer owns a native implementation

Specified by:

hasNativeImplementation in interface Renormalizer<double[][]>

Returns:

true if the native implementation exists

processInput

public boolean processInput()

Description copied from interface: Renormalizer

Know if the renormalizer can handle input (call normalizeInput has effect)

Specified by:

processInput in interface Renormalizer<double[][]>

Returns:

true if it can handle

processOutput

public boolean processOutput()

Description copied from interface: Renormalizer

Know if the renormalizer can handle output (call normalizeOutput has effect)

Specified by:

processOutput in interface Renormalizer<double[][]>

Returns:

true if it can handle

usePairShuffle

public void usePairShuffle(boolean shuffle)

Enable or disable pairwise shuffle

Parameters:

shuffle - true to enable, false to disable

shufflePair

public boolean shufflePair()

Description copied from interface: Renormalizer

Know if the measurer use pre-shuffle

Specified by:

shufflePair in interface Renormalizer<double[][]>

Returns:

true if pair will be shuffled, false otherwise

cloneState

public double[][] cloneState()

Description copied from interface: Renormalizer

Avoid concurrent modification and inconsistent results when multi-threading is enabled. This methode clone the current inner state of the renormalizer in ordre to apply modications (like removing a vector data from the renormilizer) into the cloned state of the current thread

Specified by:

cloneState in interface Renormalizer<double[][]>

Returns:

The current inner state of the renormalizer