org.encog.mathutil.error

Class ErrorCalculation

java.lang.Object

org.encog.mathutil.error.ErrorCalculation

public class ErrorCalculation
extends Object

Calculate the error of a neural network. Encog currently supports three error calculation modes. See ErrorCalculationMode for more info.

Constructor Summary

Constructors

Constructor and Description
ErrorCalculation()

Method Summary

Methods

Modifier and Type	Method and Description
double	calculate() Returns the root mean square error for a complete training set.
double	calculateESS() Calculate the error with SSE.
double	calculateMSE() Calculate the error with MSE.
double	calculateRMS() Calculate the error with RMS.
static ErrorCalculationMode	getMode() get the error calculation mode, this is static and therefore global to all Enocg training.
void	reset() Reset the error accumulation to zero.
static void	setMode(ErrorCalculationMode theMode) Set the error calculation mode, this is static and therefore global to all Enocg training.
void	updateError(double[] actual, double[] ideal, double significance) Called to update for each number that should be checked.
void	updateError(double actual, double ideal) Update the error with single values.

Methods inherited from class java.lang.Object

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

Constructor Detail

ErrorCalculation

public ErrorCalculation()

Method Detail

getMode

public static ErrorCalculationMode getMode()

get the error calculation mode, this is static and therefore global to all Enocg training. If a particular training method only supports a particular error calculation method, it may override this value. It will not change the value set here, rather the training will occur with its preferred training method. Currently the only training method that does this is Levenberg Marquardt (LMA). The default error mode for Encog is MSE.

Returns:

The current mode.

setMode

public static void setMode(ErrorCalculationMode theMode)

Set the error calculation mode, this is static and therefore global to all Enocg training. If a particular training method only supports a particular error calculation method, it may override this value. It will not change the value set here, rather the training will occur with its preferred training method. Currently the only training method that does this is Levenberg Marquardt (LMA).

Parameters:

theMode - The new mode.

calculate

public final double calculate()

Returns the root mean square error for a complete training set.

Returns:

The current error for the neural network.

calculateMSE

public final double calculateMSE()

Calculate the error with MSE.

Returns:

The current error for the neural network.

calculateESS

public final double calculateESS()

Calculate the error with SSE.

Returns:

The current error for the neural network.

calculateRMS

public final double calculateRMS()

Calculate the error with RMS.

Returns:

The current error for the neural network.

reset

public final void reset()

Reset the error accumulation to zero.

updateError

public final void updateError(double actual,
               double ideal)

Update the error with single values.

Parameters:

actual - The actual value.

ideal - The ideal value.

updateError

public final void updateError(double[] actual,
               double[] ideal,
               double significance)

Called to update for each number that should be checked.

Parameters:

actual - The actual number.

ideal - The ideal number.