com.imsl.datamining.neural

Interface QuasiNewtonTrainer.Error

All Superinterfaces:

Serializable

Enclosing class:

QuasiNewtonTrainer

public static interface QuasiNewtonTrainer.Error
extends Serializable

Error function to be minimized by trainer. This trainer attempts to solve the problem $$\min_{w} \sum_{i=0}^{n-1} e(y_i, \hat{y}_i)$$ where w are the weights, n is the number of training patterns, \(y_i\) is a training target output and \(\hat{y}_i\) is its forecast value.

This interface defines the function \(e(y, \hat{y})\) and its derivative with respect to its computed value, \(de/d\hat{y}\).

Method Summary

Modifier and Type	Method and Description
double	error(double[] computed, double[] expected) Returns the contribution to the error from a single training output target.
double[]	errorGradient(double[] computed, double[] expected) Returns the derivative of the error function with respect to the forecast output.

Method Detail

error

double error(double[] computed,
             double[] expected)

Returns the contribution to the error from a single training output target. This is the function \(e(y_i, \hat{y}_i)\).

Parameters:

computed - A double representing the computed value.

expected - A double representing the expected value.

Returns:

A double representing the contribution to the error from a single training output target.

errorGradient

double[] errorGradient(double[] computed,
                       double[] expected)

Returns the derivative of the error function with respect to the forecast output.

Parameters:

computed - A double representing the computed value.

expected - A double representing the expected value.

Returns:

A double representing the derivative of the error function with respect to the forecast output.