weka.core.metrics
Class WeightedMahalanobis

java.lang.Object
  extended byweka.core.metrics.Metric
      extended byweka.core.metrics.LearnableMetric
          extended byweka.core.metrics.WeightedMahalanobis
All Implemented Interfaces:
java.lang.Cloneable, OptionHandler, java.io.Serializable

public class WeightedMahalanobis
extends LearnableMetric
implements OptionHandler

WeightedMahalanobis class Implements a weighted Mahalanobis distance metric weighted by a full matrix of weights.

See Also:
Serialized Form

Field Summary
static int CONVERSION_EXPONENTIAL
           
static int CONVERSION_LAPLACIAN
          We can have different ways of converting from distance to similarity
static int CONVERSION_UNIT
           
protected  int m_conversionType
          The method of converting, by default laplacian
protected  double[][] m_maxPoints
          Max instance storage (max is in the space of projected instances, values are in **ORIGINAL** space) Currently somewhat convoluted, TODO: re-write the max value code in MPCKMeans
protected  double[][] m_maxProjPoints
           
protected  java.util.HashMap m_projectedInstanceHash
          A hash where instances are projected using the weights
protected  double[][] m_weights
          The full matrix of attribute weights
protected  double[][] m_weightsSquare
          weights^0.5, used to project instances to the new space to speed up calculations
static Tag[] TAGS_CONVERSION
           
 
Fields inherited from class weka.core.metrics.LearnableMetric
m_attrWeights, m_classifier, m_classifierClassName, m_classifierRequiresNominalClass, m_numPosDiffInstances, m_posNegDiffInstanceRatio, m_trainable
 
Fields inherited from class weka.core.metrics.Metric
m_attrIdxs, m_classIndex, m_numAttributes
 
Constructor Summary
WeightedMahalanobis()
          Create a default new metric
WeightedMahalanobis(int numAttributes)
          Create a new metric.
WeightedMahalanobis(int[] _attrIdxs)
          Creates a new metric which takes specified attributes.
 
Method Summary
 void buildMetric(Instances data)
          Create a new metric for operating on specified instances
 void buildMetric(int numAttributes)
          Generates a new Metric.
 void buildMetric(int numAttributes, java.lang.String[] options)
          Generates a new Metric.
 java.lang.Object clone()
          Create a copy of this metric
 Instance createDiffInstance(Instance instance1, Instance instance2)
          Create an instance with features corresponding to dot-product components of the two given instances
 Jama.Matrix createDiffMatrix(Instance instance1, Instance instance2)
          Create a matrix of the form (inst1 - inst2) * (inst1 - inst2)^T
 double distance(Instance instance1, Instance instance2)
          Returns a distance value between two instances.
 double distanceNonWeighted(Instance instance1, Instance instance2)
          Returns a non-weighted distance value between two instances.
 Instance getCentroidInstance(Instances instances, boolean fastMode, boolean normalized)
          Given a cluster of instances, return the centroid of that cluster
 SelectedTag getConversionType()
          return the type of distance to similarity conversion
 double[] getGradients(Instance instance1, Instance instance2)
          Get the values of the partial derivates for the metric components for a particular instance pair
 double[][] getMaxPoints(java.util.HashMap constraintMap, Instances instances)
          Get the maxPoints instances
 java.lang.String[] getOptions()
          Gets the current settings of WeightedMahalanobisP.
 double[] getWeights()
          override the parent class methods
 Jama.Matrix getWeightsMatrix()
          override the parent class methods
 boolean isDistanceBased()
          The computation of a metric can be either based on distance, or on similarity
 void learnMetric(Instances data)
          Train the metric
 java.util.Enumeration listOptions()
          Returns an enumeration describing the available options.
static void main(java.lang.String[] args)
           
 double[] projectInstance(Instance instance)
          given an instance, project it using the weights matrix and store it in the hash
 void resetMetric()
          Reset all values that have been learned
 void setConversionType(SelectedTag conversionType)
          Set the type of distance to similarity conversion.
 void setOptions(java.lang.String[] options)
          Parses a given list of options.
 void setWeights(double[] weights)
          override the parent class methods
 void setWeights(Jama.Matrix weights)
          Set the weights
 double similarity(Instance instance1, Instance instance2)
          Returns a similarity estimate between two instances.
 double similarityNonWeighted(Instance instance1, Instance instance2)
          Returns a similarity estimate between two instances without using the weights.
 
Methods inherited from class weka.core.metrics.LearnableMetric
getExternal, getNumPosDiffInstances, getPosNegDiffInstanceRatio, getTrainable, meanOrMode, normalizeInstanceWeighted, setExternal, setNumPosDiffInstances, setPosNegDiffInstanceRatio, setTrainable, useClassifier, useNoClassifier, usesClassifier
 
Methods inherited from class weka.core.metrics.Metric
forName, getAttrIdxs, getAttrIdxsWithoutLastClass, getAttrIndxs, getClassIndex, getNumAttributes, length, normalizeInstance, setAttrIdxs, setAttrIdxs, setClassIndex
 
Methods inherited from class java.lang.Object
equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Field Detail

m_weights

protected double[][] m_weights
The full matrix of attribute weights


m_weightsSquare

protected double[][] m_weightsSquare
weights^0.5, used to project instances to the new space to speed up calculations


m_projectedInstanceHash

protected java.util.HashMap m_projectedInstanceHash
A hash where instances are projected using the weights


m_maxPoints

protected double[][] m_maxPoints
Max instance storage (max is in the space of projected instances, values are in **ORIGINAL** space) Currently somewhat convoluted, TODO: re-write the max value code in MPCKMeans


m_maxProjPoints

protected double[][] m_maxProjPoints

CONVERSION_LAPLACIAN

public static final int CONVERSION_LAPLACIAN
We can have different ways of converting from distance to similarity

See Also:
Constant Field Values

CONVERSION_UNIT

public static final int CONVERSION_UNIT
See Also:
Constant Field Values

CONVERSION_EXPONENTIAL

public static final int CONVERSION_EXPONENTIAL
See Also:
Constant Field Values

TAGS_CONVERSION

public static final Tag[] TAGS_CONVERSION

m_conversionType

protected int m_conversionType
The method of converting, by default laplacian

Constructor Detail

WeightedMahalanobis

public WeightedMahalanobis(int numAttributes)
                    throws java.lang.Exception
Create a new metric.

Parameters:
numAttributes - the number of attributes that the metric will work on

WeightedMahalanobis

public WeightedMahalanobis()
Create a default new metric


WeightedMahalanobis

public WeightedMahalanobis(int[] _attrIdxs)
                    throws java.lang.Exception
Creates a new metric which takes specified attributes.

Parameters:
_attrIdxs - An array containing attribute indeces that will be used in the metric
Method Detail

resetMetric

public void resetMetric()
                 throws java.lang.Exception
Reset all values that have been learned

Specified by:
resetMetric in class LearnableMetric
Throws:
java.lang.Exception

buildMetric

public void buildMetric(int numAttributes)
                 throws java.lang.Exception
Generates a new Metric. Has to initialize all fields of the metric with default values.

Specified by:
buildMetric in class Metric
Parameters:
numAttributes - the number of attributes that the metric will work on
Throws:
java.lang.Exception - if the distance metric has not been generated successfully.

buildMetric

public void buildMetric(int numAttributes,
                        java.lang.String[] options)
                 throws java.lang.Exception
Generates a new Metric. Has to initialize all fields of the metric with default values

Specified by:
buildMetric in class Metric
Parameters:
options - an array of options suitable for passing to setOptions. May be null.
numAttributes - the number of attributes that the metric will work on
Throws:
java.lang.Exception - if the distance metric has not been generated successfully.

buildMetric

public void buildMetric(Instances data)
                 throws java.lang.Exception
Create a new metric for operating on specified instances

Specified by:
buildMetric in class Metric
Parameters:
data - instances that the metric will be used on
Throws:
java.lang.Exception

distance

public double distance(Instance instance1,
                       Instance instance2)
                throws java.lang.Exception
Returns a distance value between two instances.

Specified by:
distance in class Metric
Parameters:
instance1 - First instance.
instance2 - Second instance.
Throws:
java.lang.Exception - if distance could not be estimated.

projectInstance

public double[] projectInstance(Instance instance)
given an instance, project it using the weights matrix and store it in the hash


getMaxPoints

public double[][] getMaxPoints(java.util.HashMap constraintMap,
                               Instances instances)
                        throws java.lang.Exception
Get the maxPoints instances

Throws:
java.lang.Exception

distanceNonWeighted

public double distanceNonWeighted(Instance instance1,
                                  Instance instance2)
                           throws java.lang.Exception
Returns a non-weighted distance value between two instances.

Specified by:
distanceNonWeighted in class Metric
Parameters:
instance1 - First instance.
instance2 - Second instance.
Throws:
java.lang.Exception - if distance could not be estimated.

similarity

public double similarity(Instance instance1,
                         Instance instance2)
                  throws java.lang.Exception
Returns a similarity estimate between two instances. Similarity is obtained by inverting the distance value using one of three methods: CONVERSION_LAPLACIAN, CONVERSION_EXPONENTIAL, CONVERSION_UNIT.

Specified by:
similarity in class Metric
Parameters:
instance1 - First instance.
instance2 - Second instance.
Throws:
java.lang.Exception - if similarity could not be estimated.

similarityNonWeighted

public double similarityNonWeighted(Instance instance1,
                                    Instance instance2)
                             throws java.lang.Exception
Returns a similarity estimate between two instances without using the weights.

Specified by:
similarityNonWeighted in class Metric
Parameters:
instance1 - First instance.
instance2 - Second instance.
Throws:
java.lang.Exception - if similarity could not be estimated.

getGradients

public double[] getGradients(Instance instance1,
                             Instance instance2)
                      throws java.lang.Exception
Get the values of the partial derivates for the metric components for a particular instance pair

Specified by:
getGradients in class LearnableMetric
Parameters:
instance1 - the first instance
instance2 - the first instance
Throws:
java.lang.Exception

learnMetric

public void learnMetric(Instances data)
                 throws java.lang.Exception
Train the metric

Specified by:
learnMetric in class LearnableMetric
Throws:
java.lang.Exception

setWeights

public void setWeights(Jama.Matrix weights)
Set the weights


setWeights

public void setWeights(double[] weights)
override the parent class methods

Overrides:
setWeights in class LearnableMetric

getWeights

public double[] getWeights()
override the parent class methods

Overrides:
getWeights in class LearnableMetric
Returns:
an array of feature weights

getWeightsMatrix

public Jama.Matrix getWeightsMatrix()
override the parent class methods


createDiffInstance

public Instance createDiffInstance(Instance instance1,
                                   Instance instance2)
Create an instance with features corresponding to dot-product components of the two given instances

Specified by:
createDiffInstance in class LearnableMetric
Parameters:
instance1 - first instance
instance2 - second instance

createDiffMatrix

public Jama.Matrix createDiffMatrix(Instance instance1,
                                    Instance instance2)
Create a matrix of the form (inst1 - inst2) * (inst1 - inst2)^T

Parameters:
instance1 - first instance
instance2 - second instance

setConversionType

public void setConversionType(SelectedTag conversionType)
Set the type of distance to similarity conversion. Values other than CONVERSION_LAPLACIAN, CONVERSION_UNIT, or CONVERSION_EXPONENTIAL will be ignored


getConversionType

public SelectedTag getConversionType()
return the type of distance to similarity conversion

Returns:
one of CONVERSION_LAPLACIAN, CONVERSION_UNIT, or CONVERSION_EXPONENTIAL

isDistanceBased

public boolean isDistanceBased()
The computation of a metric can be either based on distance, or on similarity

Specified by:
isDistanceBased in class Metric

getCentroidInstance

public Instance getCentroidInstance(Instances instances,
                                    boolean fastMode,
                                    boolean normalized)
Given a cluster of instances, return the centroid of that cluster

Specified by:
getCentroidInstance in class LearnableMetric
Parameters:
instances - objects belonging to a cluster
fastMode - whether fast mode should be used for SparseInstances
normalized - normalize centroids for SPKMeans
Returns:
a centroid instance for the given cluster

setOptions

public void setOptions(java.lang.String[] options)
                throws java.lang.Exception
Parses a given list of options. Valid options are:

-N
Normalize the euclidean distance by vectors lengths -E
Use exponential conversion from distance to similarity (default laplacian conversion)

-U
Use unit conversion from similarity to distance (dist=1-sim) (default laplacian conversion)

-R
The metric is trainable and will be trained using the current MetricLearner (default non-trainable)

Specified by:
setOptions in interface OptionHandler
Parameters:
options - the list of options as an array of strings
Throws:
java.lang.Exception - if an option is not supported

listOptions

public java.util.Enumeration listOptions()
Returns an enumeration describing the available options.

Specified by:
listOptions in interface OptionHandler
Returns:
an enumeration of all the available options.

getOptions

public java.lang.String[] getOptions()
Gets the current settings of WeightedMahalanobisP.

Specified by:
getOptions in interface OptionHandler
Returns:
an array of strings suitable for passing to setOptions()

clone

public java.lang.Object clone()
Create a copy of this metric

Overrides:
clone in class LearnableMetric

main

public static void main(java.lang.String[] args)