org.nd4j.linalg.lossfunctions

Class LossFunctions

java.lang.Object

org.nd4j.linalg.lossfunctions.LossFunctions

public class LossFunctions
extends Object

Central class for loss functions

Author:

Adam Gibson

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	LossFunctions.LossFunction MSE: Mean Squared Error: Linear Regression EXPLL: Exponential log likelihood: Poisson Regression XENT: Cross Entropy: Binary Classification MCXENT: Multiclass Cross Entropy RMSE_XENT: RMSE Cross Entropy SQUARED_LOSS: Squared Loss NEGATIVELOGLIKELIHOOD: Negative Log Likelihood

Constructor Summary

Constructors

Constructor and Description
LossFunctions()

Method Summary

Modifier and Type	Method and Description
static double	score(INDArray labels, LossFunctions.LossFunction lossFunction, INDArray z, double l2, double l1, boolean useRegularization) Generic scoring function.

Methods inherited from class java.lang.Object

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

Constructor Detail

LossFunctions

public LossFunctions()

Method Detail

score

public static double score(INDArray labels,
                           LossFunctions.LossFunction lossFunction,
                           INDArray z,
                           double l2,
                           double l1,
                           boolean useRegularization)

Generic scoring function. Note that an IllegalArgumentException is thrown if the given loss function is custom. An alternative mechanism for scoring (preferrably with a function name and the op factory) should be used instead.

Parameters:

labels - the labels to score

lossFunction - the loss function to use

z - the output function

l2 - the l2 regularization term (0.5 * l2Coeff * sum w^2)

l1 - the l1 regularization term (l1Coeff * sum |w|)

useRegularization - whether to use regularization

Returns:

the score for the given parameters