public class LMS extends SupervisedLearning implements Serializable
maxError, previousEpochError
currentIteration, learningRate, stopConditions
listeners, neuralNetwork, trainingSet
Constructor and Description |
---|
LMS()
Creates a new LMS learning rule
|
Modifier and Type | Method and Description |
---|---|
protected void |
calculateWeightChanges(double[] outputError)
This method implements the weights update procedure for the whole network
for the given output error vector.
|
void |
updateNeuronWeights(Neuron neuron)
This method implements weights update procedure for the single neuron
It iterates through all neuron's input connections, and calculates/set weight change for each weight
using formula
deltaWeight = learningRate * neuronError * input
where neuronError is difference between desired and actual output for specific neuron
neuronError = desiredOutput[i] - actualOutput[i] (see method SuprevisedLearning.calculateOutputError)
|
afterEpoch, beforeEpoch, doBatchWeightsUpdate, doLearningEpoch, getErrorFunction, getMaxError, getMinErrorChange, getMinErrorChangeIterationsCount, getMinErrorChangeIterationsLimit, getPreviousEpochError, getTotalNetworkError, isBatchMode, learn, learn, learnPattern, onStart, setBatchMode, setErrorFunction, setMaxError, setMinErrorChange, setMinErrorChangeIterationsLimit
doOneLearningIteration, getCurrentIteration, getLearningRate, getMaxIterations, hasReachedStopCondition, isIterationsLimited, isPausedLearning, learn, learn, pause, resume, setLearningRate, setMaxIterations
addListener, fireLearningEvent, getNeuralNetwork, getTrainingSet, isStopped, onStop, removeListener, setNeuralNetwork, setTrainingSet, stopLearning
protected void calculateWeightChanges(double[] outputError)
calculateWeightChanges
in class SupervisedLearning
outputError
- output error vector for some network input- the difference between desired and actual outputlearnPattern
public void updateNeuronWeights(Neuron neuron)
neuron
- neuron to update weightscalculateWeightChanges(double[])
Copyright © 2017 Neuroph Project. All rights reserved.