jsat.classifiers.evaluation

Class LogLoss

java.lang.Object

jsat.classifiers.evaluation.LogLoss

All Implemented Interfaces:

Serializable, ClassificationScore

public class LogLoss
extends Object
implements ClassificationScore

This computes the multi-class Log Loss
- 1/N Σ_{∀ i ∈ N} log(p_{i, y})

Where N is the number of data points and p_{i, y} is the estimated probability of the true class label. The lower the loss score, the better.

When p_{i, y} = 0 the log loss is uninformatively forced to ∞, even if all other data points are perfectly correct. To avoid this a small nudge factor is added.

Author:

Edward Raff

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
LogLoss() Creates a new Log Loss evaluation score
LogLoss(double nudge) Creates a new Log Loss evaluation score
LogLoss(LogLoss toClone)

Method Summary

Modifier and Type	Method and Description
void	addResult(CategoricalResults prediction, int trueLabel, double weight) Adds the given result to the score
void	addResults(ClassificationScore other) The score contained in this object is augmented with the results already accumulated in the other object.
LogLoss	clone()
boolean	equals(Object obj)
String	getName() Returns the name to present for this score
double	getScore() Computes the score for the results that have been enrolled via ClassificationScore.addResult(jsat.classifiers.CategoricalResults, int, double)
int	hashCode()
boolean	lowerIsBetter() Returns true if a lower score is better, or false if a higher score is better
void	prepare(CategoricalData toPredict) Prepares this score to predict on the given input

Methods inherited from class java.lang.Object

finalize, getClass, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

LogLoss

public LogLoss()

Creates a new Log Loss evaluation score

LogLoss

public LogLoss(double nudge)

Creates a new Log Loss evaluation score

Parameters:

nudge - the nudge value to avoid zero probabilities, must be non negative and less than 0.1

LogLoss

public LogLoss(LogLoss toClone)

Method Detail

addResult

public void addResult(CategoricalResults prediction,
                      int trueLabel,
                      double weight)

Description copied from interface: ClassificationScore

Adds the given result to the score

Specified by:

addResult in interface ClassificationScore

Parameters:

prediction - the prediction for the data point

trueLabel - the true label for the data point

weight - the weigh to assign to the data point

addResults

public void addResults(ClassificationScore other)

Description copied from interface: ClassificationScore

The score contained in this object is augmented with the results already accumulated in the other object. This does not result in an averaging, but alters the current object to have the same score it would have had if all the results were originally inserted into this object.

This method is only required to work if other if of the same class as this object.

Specified by:

addResults in interface ClassificationScore

Parameters:

other - the object to add the results from

prepare

public void prepare(CategoricalData toPredict)

Description copied from interface: ClassificationScore

Prepares this score to predict on the given input

Specified by:

prepare in interface ClassificationScore

Parameters:

toPredict - the class label information that will be evaluated

getScore

public double getScore()

Description copied from interface: ClassificationScore

Computes the score for the results that have been enrolled via ClassificationScore.addResult(jsat.classifiers.CategoricalResults, int, double)

Specified by:

getScore in interface ClassificationScore

Returns:

the score for the current results

equals

public boolean equals(Object obj)

Specified by:

equals in interface ClassificationScore

Overrides:

equals in class Object

hashCode

public int hashCode()

Specified by:

hashCode in interface ClassificationScore

Overrides:

hashCode in class Object

lowerIsBetter

public boolean lowerIsBetter()

Description copied from interface: ClassificationScore

Returns true if a lower score is better, or false if a higher score is better

Specified by:

lowerIsBetter in interface ClassificationScore

Returns:

true if a lower score is better

clone

public LogLoss clone()

Specified by:

clone in interface ClassificationScore

Overrides:

clone in class Object

getName

public String getName()

Description copied from interface: ClassificationScore

Returns the name to present for this score

Specified by:

getName in interface ClassificationScore

Returns:

the score name