Trainer (boone)

java.lang.Object
- boone.Trainer

All Implemented Interfaces:

Storable, java.lang.Cloneable

Direct Known Subclasses:

BackpropTrainer, HebbTrainer, HopfieldDeltaTrainer, SAETrainer, SOMTrainer
```
public abstract class Trainer
extends java.lang.Object
implements java.lang.Cloneable, Storable
```
The trainer super class for all training algorithms.

Version:

$Id$

Author:

August Mayer, Helmut A. Mayer

Field Summary

Fields
Modifier and Type	Field and Description
`protected int`	`currentEpoch` The current epoch during training.
`protected int`	`epochs` The number of epochs to be trained.
`protected double`	`learnRate` The learn rate of this trainer.
`protected NeuralNet`	`net` The network to be trained.
`protected boolean`	`shuffle` Indicates random shuffling of training data.
`protected boolean`	`stepMode` Indicates step-wise training.
`protected PatternSet`	`testData`
`protected PatternSet`	`trainingData` The training data set.
`protected TrainingSignalGenerator`	`trainingSignalGenerator` The training signal generator.

Constructor Summary

Constructors
Constructor and Description

Trainer()
Constructs a trainer with squared error function (learn rate = 0.2, epochs = 1000).

Constructors
Constructor and Description
`Trainer()` Constructs a trainer with squared error function (learn rate = 0.2, epochs = 1000).

Method Summary

All Methods Instance Methods Abstract Methods Concrete Methods
Modifier and Type	Method and Description
`protected abstract double`	`calcAdaptation(BrainPart part)` Calculates the adaptation value based on the gradient.
`Trainer`	`clone()` Return a deep clone of this object.
`protected void`	`endBatch()` Updates weights and biases using the batch gradient information.
`protected abstract void`	`endTrain()` Do some house keeping after all epochs have been trained.
`void`	`fromXML(org.jdom2.Element element)` Reads attributes from XML.
`int`	`getBatchSize()` Returns the (mini)batch size.
`int`	`getEpochs()` Returns the number of epochs to be trained.
`double`	`getLearnRate()` Returns the learning rate of the trainer.
`NeuralNet`	`getNetwork()` Returns the network associated with this trainer.
`int`	`getRank(java.util.List<java.lang.Double> input, int index)` Returns the ranked activation of the given output neuron and the given input pattern.
`PatternSet`	`getTestData()` Returns the programs data set.
`PatternSet`	`getTrainingData()` Returns the training data set.
`TrainingSignalGenerator`	`getTrainingSignalGenerator()` Returns the current training signal generator for this trainer.
`Neuron`	`getWinningNeuron(java.util.List<java.lang.Double> input)` Returns the output neuron with largest activation upon presentation of 'input'.
`protected abstract void`	`reset()` Resets the trainer to initial values in order to start a new training procedure.
`void`	`setBatchSize(int batchSize)` Sets the (mini)batch size.
`void`	`setEpochs(int epochs)` Sets the number of epochs.
`void`	`setLearnRate(double learnRate)` Set the learn rate for this Trainer.
`void`	`setNetwork(NeuralNet net)` Sets the network for this trainer.
`void`	`setShuffle(boolean shuffle)` Sets the shuffle flag.
`void`	`setStepMode(boolean stepMode)` Sets the step mode.
`void`	`setTestData(PatternSet testData)` Sets the programs data set.
`void`	`setTrainingData(PatternSet trainingData)` Sets the training data set.
`void`	`setTrainingSignalGenerator(TrainingSignalGenerator trainingSignalGenerator)` Set the training signal generator for this trainer.
`double`	`test()` Returns the error on the test data.
`double`	`test(java.util.List<java.lang.Double> input, java.util.List<java.lang.Double> target)` Test a given input pattern and return the error.
`java.lang.String`	`toString()`
`org.jdom2.Element`	`toXML(org.jdom2.Element element)` Writes attributes to XML.
`void`	`train()` Performs training in the following order:
`protected abstract void`	`train(java.util.List<java.lang.Double> inputPattern, java.util.List<java.lang.Double> targetPattern)` Trains a single pattern once.
`void`	`trainPattern(java.util.List<java.lang.Double> inputPattern, java.util.List<java.lang.Double> targetPattern)` Trains a single pattern once and updates the weights.
`protected void`	`updateLinks()` Adapts neuron link values after a (mini)batch.
`protected void`	`updateNeurons()` Adapts neuron bias values after a (mini)batch.

Methods inherited from class java.lang.Object
equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

- Field Detail
  - net
```
protected NeuralNet net
```
    The network to be trained.
  - trainingData
```
protected PatternSet trainingData
```
    The training data set.
  - testData
```
protected PatternSet testData
```
  - trainingSignalGenerator
```
protected TrainingSignalGenerator trainingSignalGenerator
```
    The training signal generator.
  - stepMode
```
protected boolean stepMode
```
    Indicates step-wise training.
  - shuffle
```
protected boolean shuffle
```
    Indicates random shuffling of training data.
  - learnRate
```
protected double learnRate
```
    The learn rate of this trainer.
  - epochs
```
protected int epochs
```
    The number of epochs to be trained.
  - currentEpoch
```
protected int currentEpoch
```
    The current epoch during training.
- Constructor Detail
  - Trainer
```
public Trainer()
```
    Constructs a trainer with squared error function (learn rate = 0.2, epochs = 1000).
- Method Detail
  - clone
```
public Trainer clone()
```
    Return a deep clone of this object. The clone has set the Network reference to null; this needs to be initialized. The name is taken unchanged from the original object; we assign a new ID, however.
    
    Overrides:
    
    clone in class java.lang.Object
  - getTrainingData
```
public PatternSet getTrainingData()
```
    Returns the training data set.
    
    Returns:
    
    a pattern set
  - setTrainingData
```
public void setTrainingData(PatternSet trainingData)
```
    Sets the training data set. Also, sets the (mini)batch size to the complete batch size, if the size has not been set.
    
    Parameters:
    
    trainingData - a pattern set (its size is the default batch size)
  - getTestData
```
public PatternSet getTestData()
```
    Returns the programs data set.
    
    Returns:
    
    a pattern set
  - setTestData
```
public void setTestData(PatternSet testData)
```
    Sets the programs data set.
    
    Parameters:
    
    testData - a pattern set
  - getBatchSize
```
public int getBatchSize()
```
    Returns the (mini)batch size.
    
    Returns:
    
    the size of a (mini)batch
  - setBatchSize
```
public void setBatchSize(int batchSize)
```
    Sets the (mini)batch size.
    
    Parameters:
    
    batchSize - the size of a (mini)batch
  - setStepMode
```
public void setStepMode(boolean stepMode)
```
    Sets the step mode. If set, the trainer assumes that training is done in steps of the number of epochs set. Hence, a next call to train() continues the training, if step mode is set. If it is not set, the next call to train() is a new training procedure.
    
    Parameters:
    
    stepMode - the mode flag
  - setShuffle
```
public void setShuffle(boolean shuffle)
```
    Sets the shuffle flag. If set, the training set is shuffled after each epoch.
    
    Parameters:
    
    shuffle - if true, shuffle training data
  - train
```
public void train()
```
    Performs training in the following order:
    - calls reset(), if not step mode - runs through the number of epochs presenting each pattern by calling train(java.util.List, java.util.List) - after each epoch endBatch() is called, then training data are shuffled (setShuffle(boolean)) - after all epochs are done, endTrain() is called
    Also works for streamed training patterns.
  - trainPattern
```
public void trainPattern(java.util.List<java.lang.Double> inputPattern,
                         java.util.List<java.lang.Double> targetPattern)
```
    Trains a single pattern once and updates the weights. This is just a delegation to internal training.
    
    Parameters:
    
    inputPattern - an input pattern
    
    targetPattern - a target pattern (may be null)
  - test
```
public double test(java.util.List<java.lang.Double> input,
                   java.util.List<java.lang.Double> target)
```
    Test a given input pattern and return the error.
    
    Parameters:
    
    input - input pattern
    
    target - target pattern
    
    Returns:
    
    test error (as provided by the TrainingSignalGenerator)
  - test
```
public double test()
```
    Returns the error on the test data. Also works for streamed test data.
    
    Returns:
    
    sum of errors per test pattern (computed by TrainingSignalGenerator)
  - endBatch
```
protected void endBatch()
```
    Updates weights and biases using the batch gradient information.
  - updateNeurons
```
protected void updateNeurons()
```
    Adapts neuron bias values after a (mini)batch.
  - updateLinks
```
protected void updateLinks()
```
    Adapts neuron link values after a (mini)batch.
  - getNetwork
```
public NeuralNet getNetwork()
```
    Returns the network associated with this trainer.
    
    Returns:
    
    the net
  - setNetwork
```
public void setNetwork(NeuralNet net)
```
    Sets the network for this trainer.
    
    Parameters:
    
    net - a neural net
  - setLearnRate
```
public void setLearnRate(double learnRate)
```
    Set the learn rate for this Trainer.
    
    Parameters:
    
    learnRate - the learning rate
    
    See Also:
    
    learnRate
  - getLearnRate
```
public double getLearnRate()
```
    Returns the learning rate of the trainer.
    
    Returns:
    
    the learning rate
    
    See Also:
    
    learnRate
  - getEpochs
```
public int getEpochs()
```
    Returns the number of epochs to be trained.
    
    Returns:
    
    the number of epochs
    
    See Also:
    
    epochs
  - setEpochs
```
public void setEpochs(int epochs)
```
    Sets the number of epochs. An epoch is a single presentation of all training patterns.
    
    Parameters:
    
    epochs - the number of epochs
    
    See Also:
    
    epochs
  - getTrainingSignalGenerator
```
public TrainingSignalGenerator getTrainingSignalGenerator()
```
    Returns the current training signal generator for this trainer. A trainer may or may not use this generator, depending on the implementation.
  - setTrainingSignalGenerator
```
public void setTrainingSignalGenerator(TrainingSignalGenerator trainingSignalGenerator)
```
    Set the training signal generator for this trainer. A trainer may or may not use this generator, depending on the implementation.
  - getWinningNeuron
```
public Neuron getWinningNeuron(java.util.List<java.lang.Double> input)
```
    Returns the output neuron with largest activation upon presentation of 'input'.
    
    Parameters:
    
    input - an input pattern
    
    Returns:
    
    the winning neuron
  - getRank
```
public int getRank(java.util.List<java.lang.Double> input,
                   int index)
```
    Returns the ranked activation of the given output neuron and the given input pattern. A rank of 1 indicates a winning neuron. In case of a tie, the given neuron receives the lowest rank of the tied neurons.
    
    Parameters:
    
    input - an input pattern
    
    index - the index of the output neuron
    
    Returns:
    
    the rank of the output neuron
  - toString
```
public java.lang.String toString()
```
    Overrides:
    
    toString in class java.lang.Object
  - fromXML
```
public void fromXML(org.jdom2.Element element)
```
    Reads attributes from XML.
    
    Specified by:
    
    fromXML in interface Storable
    
    Parameters:
    
    element - a JDOM element
  - toXML
```
public org.jdom2.Element toXML(org.jdom2.Element element)
```
    Writes attributes to XML.
    
    Specified by:
    
    toXML in interface Storable
    
    Parameters:
    
    element - the parent element
    
    Returns:
    
    the trainer element
  - reset
```
protected abstract void reset()
```
    Resets the trainer to initial values in order to start a new training procedure.
  - endTrain
```
protected abstract void endTrain()
```
    Do some house keeping after all epochs have been trained.
  - train
```
protected abstract void train(java.util.List<java.lang.Double> inputPattern,
                              java.util.List<java.lang.Double> targetPattern)
```
    Trains a single pattern once.
    
    Parameters:
    
    inputPattern - an input pattern
    
    targetPattern - a target pattern (may be null)
  - calcAdaptation
```
protected abstract double calcAdaptation(BrainPart part)
```
    Calculates the adaptation value based on the gradient.
    
    Parameters:
    
    part - the brain part to be adapted
    
    Returns:
    
    the adaptation value

Class Trainer

Field Summary

Constructor Summary

Method Summary

Methods inherited from class java.lang.Object

Field Detail

net

trainingData

testData

trainingSignalGenerator

stepMode

shuffle

learnRate

epochs

currentEpoch

Constructor Detail

Trainer

Method Detail

clone

getTrainingData

setTrainingData

getTestData

setTestData

getBatchSize

setBatchSize

setStepMode

setShuffle

train

trainPattern

test

test

endBatch

updateNeurons

updateLinks

getNetwork

setNetwork

setLearnRate

getLearnRate

getEpochs

setEpochs

getTrainingSignalGenerator

setTrainingSignalGenerator

getWinningNeuron

getRank

toString

fromXML

toXML

reset

endTrain

train

calcAdaptation