Package de.jungblut.math.loss

Interface LossFunction

All Known Implementing Classes:

CrossEntropyLoss, HingeLoss, LogLoss, MeanAbsoluteLoss, SquaredLoss, StepLoss

public interface LossFunction

Calculates the error, for example in the last layer of a neural net.

Author:

thomas.jungblut

Method Summary

Modifier and Type	Method	Description
de.jungblut.math.DoubleVector	calculateGradient(de.jungblut.math.DoubleVector feature, de.jungblut.math.DoubleVector y, de.jungblut.math.DoubleVector hypothesis)	Calculate the gradient with the given parameters.
double	calculateLoss(de.jungblut.math.DoubleMatrix y, de.jungblut.math.DoubleMatrix hypothesis)	Calculate the error with the given parameters.
double	calculateLoss(de.jungblut.math.DoubleVector y, de.jungblut.math.DoubleVector hypothesis)	Calculate the error with the given parameters.

Method Detail

calculateLoss

double calculateLoss(de.jungblut.math.DoubleMatrix y,
                     de.jungblut.math.DoubleMatrix hypothesis)

Calculate the error with the given parameters.

Parameters:

y - the real outcome as a matrix- rows contain the examples, columns the examples' output.

hypothesis - the hypothesis as a matrix- rows contain the examples, columns the predicted output.

Returns:

a positive value that denotes the error between the two matrices.

calculateLoss

double calculateLoss(de.jungblut.math.DoubleVector y,
                     de.jungblut.math.DoubleVector hypothesis)

Calculate the error with the given parameters.

Parameters:

y - the real outcome as a vector single example.

hypothesis - the hypothesis as a vector single example.

Returns:

a positive value that denotes the error between the two vectors.

calculateGradient

de.jungblut.math.DoubleVector calculateGradient(de.jungblut.math.DoubleVector feature,
                                                de.jungblut.math.DoubleVector y,
                                                de.jungblut.math.DoubleVector hypothesis)

Calculate the gradient with the given parameters.

Parameters:

y - the real outcome as a vector single example.

hypothesis - the hypothesis as a vector single example.

Returns:

a vector that denotes the gradient given the hypothesis and real outcome.