mulan.data
Class ConditionalDependenceIdentifier

java.lang.Object
  mulan.data.ConditionalDependenceIdentifier

All Implemented Interfaces:

Serializable, LabelPairsDependenceIdentifier

public class ConditionalDependenceIdentifier
extends Object
implements LabelPairsDependenceIdentifier, Serializable

A class for identification of conditional dependence between each pair of labels. The conditional dependence between each pair of labels is estimated by evaluating the advantage gained from exploiting this dependence for binary classification of each one of the labels. Following the definition of conditional independence, for two conditionally independent labels, predictions of a label by probability-based classification models trained once on a regular features space and second on the features space augmented by the second label should be at least very similar. For this estimation two binary classifiers are trained and their accuracy is estimated using k-fold cross-validation. If the accuracy of the model trained on the features space augmented by the second label is significantly higher, the labels are considered conditionally dependent. The statistical significance of the difference between both classifiers is determined using a paired t-test. This procedure is performed for all possible label pairs considering the label order in the pair . Among the two pairs with the same labels, the pair with maximal t-statistic value is added to the resulting list of dependent pairs. Finally, the resultant label pairs are sorted according to their t-statistic value in descending order (i.e., from the most to the least dependent pairs).

Version:

30.11.2010

Author:

Lena Chekina (lenat@bgu.ac.il)

See Also:

Serialized Form

Field Summary

protected int

seed Seed for replication of random experiments

Constructor Summary

ConditionalDependenceIdentifier(Classifier classifier)
Initializes a single-label classifier used to perform dependence test between labels and a caching mechanism for reusing constructed models.

Method Summary

LabelsPair[]	calculateDependence(MultiLabelInstances mlInstances) Calculates t-statistic value for each pair of labels.
double	getCriticalValue() Returns a critical value
int	getNumFolds()
int	getSeed()
void	setCriticalValue(double criticalValue)
void	setNumFolds(int numFolds)
void	setSeed(int seed)

Methods inherited from class java.lang.Object

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

Field Detail

seed

protected int seed

Seed for replication of random experiments

Constructor Detail

ConditionalDependenceIdentifier

public ConditionalDependenceIdentifier(Classifier classifier)

Initializes a single-label classifier used to perform dependence test between labels and a caching mechanism for reusing constructed models.

Parameters:

classifier - - a single-label classifier used to perform dependence test between labels.

Method Detail

calculateDependence

public LabelsPair[] calculateDependence(MultiLabelInstances mlInstances)

Calculates t-statistic value for each pair of labels.

Specified by:

calculateDependence in interface LabelPairsDependenceIdentifier

Parameters:

mlInstances - the MultiLabelInstances dataset on which dependencies should be calculated

Returns:

an array of label pairs sorted in descending order of the t-statistic value

setCriticalValue

public void setCriticalValue(double criticalValue)

Parameters:

criticalValue -

getCriticalValue

public double getCriticalValue()

Description copied from interface: LabelPairsDependenceIdentifier

Returns a critical value

Specified by:

getCriticalValue in interface LabelPairsDependenceIdentifier

Returns:

critical value

getSeed

public int getSeed()

Returns:

setSeed

public void setSeed(int seed)

Parameters:

seed -

getNumFolds

public int getNumFolds()

Returns:

setNumFolds

public void setNumFolds(int numFolds)

Parameters:

numFolds -