de.jungblut.online.regularization

Class L2Regularizer

java.lang.Object

de.jungblut.online.regularization.GradientDescentUpdater

de.jungblut.online.regularization.L2Regularizer

All Implemented Interfaces:

WeightUpdater

public class L2Regularizer
extends GradientDescentUpdater

Computes the L2 regularized update: R(w) = (||w||^2) / 2. It assumes the bias feature to be on the very first dimension (zero index) in order to deliberately not regularize it.

Author:

thomas.jungblut

Constructor Summary

Constructors

Constructor and Description
L2Regularizer()

Method Summary

Modifier and Type	Method and Description
de.jungblut.math.minimize.CostGradientTuple	computeGradient(de.jungblut.math.DoubleVector weights, de.jungblut.math.DoubleVector gradient, double learningRate, long iteration, double lambda, double cost) Computes the gradient.

Methods inherited from class de.jungblut.online.regularization.GradientDescentUpdater

computeNewWeights

Methods inherited from class java.lang.Object

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

Constructor Detail

L2Regularizer

public L2Regularizer()

Method Detail

computeGradient

public de.jungblut.math.minimize.CostGradientTuple computeGradient(de.jungblut.math.DoubleVector weights,
                                                                   de.jungblut.math.DoubleVector gradient,
                                                                   double learningRate,
                                                                   long iteration,
                                                                   double lambda,
                                                                   double cost)

Description copied from interface: WeightUpdater

Computes the gradient.

Specified by:

computeGradient in interface WeightUpdater

Overrides:

computeGradient in class GradientDescentUpdater

Parameters:

weights - the old weights.

gradient - the gradient.

learningRate - the learning rate.

iteration - the number of the current iteration.

lambda - the regularization parameter.

cost - the computed cost for this gradient update.

Returns:

the gradient vector that should be substracted from the weights and the updated cost.