de.htwdd.rosenkoenig.neuro.net.training
Class Backpropagation

java.lang.Object
  de.htwdd.rosenkoenig.neuro.net.training.TeachingAlgorithm
      de.htwdd.rosenkoenig.neuro.net.training.Backpropagation

public class Backpropagation
extends TeachingAlgorithm

Implementation of the backpropagation teaching algorithm for feed forward neural networks.

Backpropagation performs training by spreading the error caused by a pattern uniformly on the whole net, depending on each neuron's error ratio. To calculate a neuron's error ratio, the method of decreasing gradients is used. Further information on Backpropagation can be found here.

Field Summary

Fields inherited from class de.htwdd.rosenkoenig.neuro.net.training.TeachingAlgorithm

trainer

Constructor Summary

Backpropagation()
Creates and initializes a new instance of Backpropagation.

Method Summary

void	teachPatterns(java.util.List<Pattern> patterns) Runs the backpropagation algorithm (for one cycle) on the provided pattern set.

Methods inherited from class de.htwdd.rosenkoenig.neuro.net.training.TeachingAlgorithm

getTrainer, setTrainer

Methods inherited from class java.lang.Object

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

Constructor Detail

Backpropagation

public Backpropagation()

Creates and initializes a new instance of Backpropagation.

Method Detail

teachPatterns

public final void teachPatterns(java.util.List<Pattern> patterns)

Runs the backpropagation algorithm (for one cycle) on the provided pattern set.

Steps:

• propagate pattern
• compute gradients for the output layer
• compute gradients for the hidden layers
• compute weight adjustments
• adjust the weights if (step % batchsize == 0)

Specified by:

teachPatterns in class TeachingAlgorithm

Parameters:

patterns - set of training patterns