Package deepnetts.net.loss

Class MeanSquaredErrorLoss

java.lang.Object

deepnetts.net.loss.MeanSquaredErrorLoss

All Implemented Interfaces:

LossFunction, Serializable

public final class MeanSquaredErrorLoss
extends Object
implements LossFunction, Serializable

Mean Squared Error Loss function. Sum squared errors over all input patterns and all outputs. Should be used for regression problems. Math formula: N K E = 1/(2*N*K) * SUM(SUM(y-t)^2) + regSum where N is number of patterns and K is dimension of output vector, and regSum is L1 or L2 regularization multiplied with lambda. Bishop, pg. 89, eq. 3.34 Also recommended this formula in Proben1 Technical report

See Also:

• LossFunction
• CrossEntropyLoss
• Serialized Form

Constructor Summary

Constructors

Constructor	Description
MeanSquaredErrorLoss(int layerWidth)
MeanSquaredErrorLoss(NeuralNetwork neuralNet)	Creates a new mean squared error loss for the given neural network.

Method Summary

Modifier and Type	Method	Description
float[]	addPatternError(float[] predictedOutput, float[] targetOutput)	Adds output error vector for the given predicted and target output vectors to total error sum and returns and error vector.
TensorBase	addPatternError(TensorBase predictedOut, TensorBase targetOut)
void	addRegularizationSum(float regSum)	Add regularization sum to total loss
float	getPatternLoss()
float	getTotal()	Returns the total error calculated by this loss function.
void	reset()	Resets the total error and pattern counter.

Methods inherited from class java.lang.Object

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

Methods inherited from interface deepnetts.net.loss.LossFunction

valueFor

Constructor Details

MeanSquaredErrorLoss

public MeanSquaredErrorLoss(NeuralNetwork neuralNet)

Creates a new mean squared error loss for the given neural network.

Parameters:

neuralNet -

MeanSquaredErrorLoss

public MeanSquaredErrorLoss(int layerWidth)

Method Details

addPatternError

public float[] addPatternError(float[] predictedOutput,
 float[] targetOutput)

Adds output error vector for the given predicted and target output vectors to total error sum and returns and error vector.

Specified by:

addPatternError in interface LossFunction

Parameters:

predictedOutput -

targetOutput -

Returns:

addPatternError

public TensorBase addPatternError(TensorBase predictedOut,
 TensorBase targetOut)

Specified by:

addPatternError in interface LossFunction

addRegularizationSum

public void addRegularizationSum(float regSum)

Add regularization sum to total loss

Specified by:

addRegularizationSum in interface LossFunction

Parameters:

regSum - regularization sum

getTotal

public float getTotal()

Description copied from interface: LossFunction

Returns the total error calculated by this loss function.

Specified by:

getTotal in interface LossFunction

Returns:

total error calculated by this loss function

reset

public void reset()

Description copied from interface: LossFunction

Resets the total error and pattern counter.

Specified by:

reset in interface LossFunction

getPatternLoss

public float getPatternLoss()

Specified by:

getPatternLoss in interface LossFunction