NeuralNet (boone)

java.lang.Object
- boone.NeuralNet

All Implemented Interfaces:

Storable, java.lang.Cloneable

Direct Known Subclasses:

SpikingNeuralNet
```
public class NeuralNet
extends java.lang.Object
implements Storable, java.lang.Cloneable
```
This class holds the neurons, links, trainers and other things of a neural network.
Use NetFactory to create NeuralNets conveniently.

Version:

$Id$

Author:

August Mayer, Helmuit A. Mayer

Field Summary

Fields
Modifier and Type	Field and Description
`static int`	`FEEDFORWARD` A feed-forward network.
`protected IOFilter`	`filter` The IO filter.
`protected NeuronList`	`inputNeurons` The input neurons.
`protected java.util.List<Layer>`	`layers` The ordered layers of the network.
`protected java.util.ArrayList<Link>`	`links` All the links of the network.
`protected int`	`maxCycles` The maximal number of cycles
`protected NeuronList`	`neurons` All the neurons of the network.
`protected NeuronList`	`outputNeurons` The output neurons.
`java.util.Properties`	`props` Some custom properties.
`static int`	`RECURRENT` A recurrent network.
`protected int`	`topology` The network topology.
`protected Trainer`	`trainer` The trainer for this network.
`static int`	`UNKNOWN` The topology is not determined.
`protected boolean`	`zeroInputAfterFirstCycle` Reset the external input of the input neurons to zero after the first cycle when innervating the network? Probably only useful for recurrent networks, where the internal per-neuron feed-back should take over (see `Neuron.loopOutputToInput`).

Constructor Summary

Constructors
Constructor and Description

NeuralNet()
Constructs the net with a default Boone IO filter.

Constructors
Constructor and Description
`NeuralNet()` Constructs the net with a default Boone IO filter.

Method Summary

All Methods Static Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`void`	`addLayer(Layer layer)` Adds the layer, the neurons of the layer and their input links to this net.
`void`	`addLink(Link link)` Adds a link.
`void`	`addNeuron(Neuron neuron)` Add a neuron to our list.
`NeuralNet`	`clone()` Clone method for this NeuralNet.
`void`	`fromXML(org.jdom2.Element element)` Loads the network attributes from XML.
`IOFilter`	`getFilter()` Returns the IO filter.
`int`	`getHiddenNeuronCount()` Return the number of hidden neurons, that is, neurons which are neither input nor output neurons.
`Neuron`	`getInputNeuron(int num)`
`int`	`getInputNeuronCount()` Returns the number of input neurons.
`Link`	`getLink(int num)` Returns a specific link in the net.
`int`	`getLinkCount()`
`Link`	`getLinkID(long id)` Returns the link with the given ID.
`int`	`getMaxCycles()`
`Neuron`	`getNeuron(int num)`
`int`	`getNeuronCount()` Returns the number of neurons in this net.
`Neuron`	`getNeuronID(long id)` Returns the neuron with the given ID.
`int`	`getNeuronIndex(Neuron neuron)`
`double[]`	`getOutput(double[] outputs)` Extract the output pattern from the network.
`Neuron`	`getOutputNeuron(int index)` Returns a specific output neuron in the network.
`int`	`getOutputNeuronCount()`
`int`	`getOutputNeuronIndex(Neuron output)` Returns the index of the given output neuron in the internal list.
`int`	`getTopology()` Returns the topology of the net.
`Trainer`	`getTrainer()` Returns the trainer of the net.
`boolean`	`getZeroInputAfterFirstCycle()` Return whether the NeuralNet zeroes out the external input of the input neurons after the first evaluation cycle.
`void`	`innervate()` Innervate the network, i.e.
`void`	`linkNet(NeuralNet other)` Links another net to this net (by appending at end).
`static NeuralNet`	`load(java.io.File file, IOFilter filter)` Loads a NeuralNet from the given file using the set file filter.
`void`	`randomize()` Randomizes the network parameters to some meaningful values.
`boolean`	`removeLink(Link link)` remove a link from our network.
`boolean`	`removeNeuron(Neuron neuron)` Removes a neuron and updates the associated links, but no link lists.
`void`	`replaceLink(Link oldLink, Link newLink)` replace a link with another.
`void`	`replaceNeuron(Neuron oldNeuron, Neuron newNeuron)` Replace a neuron with another.
`void`	`save(java.io.File file)` Save a NeuralNet to the given file, in the Boone XNet format, using the IOFilter class.
`void`	`setFilter(IOFilter filter)` Sets the IO filter.
`void`	`setInput(double[] values)` Load the input pattern into the network.
`void`	`setInput(java.util.List<java.lang.Double> values)` Load the input pattern into the network.
`void`	`setMaxCycles(int maxCycles)`
`void`	`setTopology(int topology)` Sets the topology of the net.
`void`	`setTrainer(Trainer trainer)` Assigns a trainer to the network and prepares the network for training.
`void`	`setZeroInputAfterFirstCycle(boolean makeZero)` Define whether to reset the external input of the input neurons to zero after the first cycle when innervating the network.
`java.lang.String`	`toString()` convert the net to a string (multi-line).
`org.jdom2.Element`	`toXML(org.jdom2.Element element)` Writes attributes to XML.

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

- Field Detail
  - UNKNOWN
```
public static final int UNKNOWN
```
    The topology is not determined.
    
    See Also:
    
    Constant Field Values
  - FEEDFORWARD
```
public static final int FEEDFORWARD
```
    A feed-forward network.
    
    See Also:
    
    Constant Field Values
  - RECURRENT
```
public static final int RECURRENT
```
    A recurrent network.
    
    See Also:
    
    Constant Field Values
  - topology
```
protected int topology
```
    The network topology.
  - layers
```
protected java.util.List<Layer> layers
```
    The ordered layers of the network.
  - neurons
```
protected NeuronList neurons
```
    All the neurons of the network.
  - inputNeurons
```
protected NeuronList inputNeurons
```
    The input neurons.
  - outputNeurons
```
protected NeuronList outputNeurons
```
    The output neurons.
  - links
```
protected java.util.ArrayList<Link> links
```
    All the links of the network.
  - trainer
```
protected Trainer trainer
```
    The trainer for this network.
  - maxCycles
```
protected int maxCycles
```
    The maximal number of cycles
  - zeroInputAfterFirstCycle
```
protected boolean zeroInputAfterFirstCycle
```
    Reset the external input of the input neurons to zero after the first cycle when innervating the network? Probably only useful for recurrent networks, where the internal per-neuron feed-back should take over (see Neuron.loopOutputToInput).
    Default: false.
  - filter
```
protected IOFilter filter
```
    The IO filter.
  - props
```
public java.util.Properties props
```
    Some custom properties.
- Constructor Detail
  - NeuralNet
```
public NeuralNet()
```
    Constructs the net with a default Boone IO filter.
- Method Detail
  - getTopology
```
public int getTopology()
```
    Returns the topology of the net.
    
    Returns:
    
    the topology value
  - setTopology
```
public void setTopology(int topology)
```
    Sets the topology of the net. The layers MUST be correctly numbered. Set it to UNKNOWN, if you are not sure.
    
    Parameters:
    
    topology - the topology value
  - getFilter
```
public IOFilter getFilter()
```
    Returns the IO filter.
    
    Returns:
    
    the IO filter
  - setFilter
```
public void setFilter(IOFilter filter)
```
    Sets the IO filter.
    
    Parameters:
    
    filter - the IO filter
  - getNeuronCount
```
public int getNeuronCount()
```
    Returns the number of neurons in this net.
    
    Returns:
    
    the number of neurons
  - getNeuron
```
public Neuron getNeuron(int num)
```
    Returns:
    
    a specific neuron in the network.
  - getNeuronIndex
```
public int getNeuronIndex(Neuron neuron)
```
    Returns:
    
    the position of the given neuron in the array, or -1.
  - getNeuronID
```
public Neuron getNeuronID(long id)
```
    Returns the neuron with the given ID.
    
    Parameters:
    
    id - the neuron ID
    
    Returns:
    
    the neuron, null if not found
  - getInputNeuronCount
```
public int getInputNeuronCount()
```
    Returns the number of input neurons.
    
    Returns:
    
    the number
  - getInputNeuron
```
public Neuron getInputNeuron(int num)
```
    Returns:
    
    a specific input neuron in the network.
  - getOutputNeuronCount
```
public int getOutputNeuronCount()
```
    Returns:
    
    the number of output neurons in the network.
  - getOutputNeuron
```
public Neuron getOutputNeuron(int index)
```
    Returns a specific output neuron in the network.
    
    Parameters:
    
    index - the index
    
    Returns:
    
    the output neuron
  - getOutputNeuronIndex
```
public int getOutputNeuronIndex(Neuron output)
```
    Returns the index of the given output neuron in the internal list. So the first output neuron has index 0.
    
    Parameters:
    
    output - an output neuron
    
    Returns:
    
    the index, -1 if the neuron is not in the list
  - getHiddenNeuronCount
```
public int getHiddenNeuronCount()
```
    Return the number of hidden neurons, that is, neurons which are neither input nor output neurons.
  - addNeuron
```
public void addNeuron(Neuron neuron)
```
    Add a neuron to our list. Specifically, adds the neuron to the internal neuron list, and to the input and/or output neuron lists if the neuron is an input/output neuron.
    
    Parameters:
    
    neuron - the neuron to add
  - addLayer
```
public void addLayer(Layer layer)
```
    Adds the layer, the neurons of the layer and their input links to this net. The layers must be added orderly starting with the input layer.
    
    Parameters:
    
    layer - the layer
  - replaceNeuron
```
public void replaceNeuron(Neuron oldNeuron,
                          Neuron newNeuron)
```
    Replace a neuron with another. The neuron is also disconnected and re-connected in all links. All links of the old neuron are added to the new neuron. If the old neuron is the same as the new neuron (w.r.t. ==), does nothing.
  - removeNeuron
```
public boolean removeNeuron(Neuron neuron)
```
    Removes a neuron and updates the associated links, but no link lists.
    
    Returns:
    
    true, if neuron removed
  - getLinkCount
```
public int getLinkCount()
```
    Returns:
    
    the number of links in the network.
  - getLink
```
public Link getLink(int num)
```
    Returns a specific link in the net. The number is just an index from 0 to #links - 1. The links are not guaranteed to be in any specific order.
    
    Returns:
    
    a specific link in the network
  - getLinkID
```
public Link getLinkID(long id)
```
    Returns the link with the given ID.
    
    Parameters:
    
    id - the link ID
    
    Returns:
    
    the link, null if not found
  - addLink
```
public void addLink(Link link)
```
    Adds a link.
    
    Parameters:
    
    link - a link
  - replaceLink
```
public void replaceLink(Link oldLink,
                        Link newLink)
```
    replace a link with another. The neuron is also disconnected and re-connected in all links. Finally, the old neuron's outputLinks and inputLinks lists are cleared.
    
    NOTE: The solution may look a bit complicated, but this is prepared for Special, extended links.
  - removeLink
```
public boolean removeLink(Link link)
```
    remove a link from our network.
    NOTE: The Link is always disconnected before removing from the neurons using Link.disconnect() .
    
    Also please note that there is no removeLink(int index) method, because the link needs to be looked up anyway, to disconnect it.
    
    Returns:
    
    whether the link was found in our list of links.
    
    See Also:
    
    Link.disconnect()
  - setInput
```
public void setInput(double[] values)
```
    Load the input pattern into the network.
  - setInput
```
public void setInput(java.util.List<java.lang.Double> values)
```
    Load the input pattern into the network.
  - getOutput
```
public double[] getOutput(double[] outputs)
```
    Extract the output pattern from the network.
    
    Parameters:
    
    outputs - Array to put the outputs into. If too small or null, we allocate a new double array for the values.
  - randomize
```
public void randomize()
```
    Randomizes the network parameters to some meaningful values.
  - innervate
```
public void innervate()
```
    Innervate the network, i.e. tick() all the neurons of the net.
    This method works as follows:
    - Check the topology of the network, if it is UNKNOWN.
    - Activate the network once, if topology is feed-forward.
    - Activate the network maxCycles times, if topology is recurrent.
    - Reset the neuron link input values to zero.
  - toString
```
public java.lang.String toString()
```
    convert the net to a string (multi-line).
    
    Overrides:
    
    toString in class java.lang.Object
  - setTrainer
```
public void setTrainer(Trainer trainer)
```
    Assigns a trainer to the network and prepares the network for training. Note that any subsequent changes to the network (by the user) may result in an incorrect training. If you have to change the network, then you must set the trainer again in order to achieve a correct training regime.
    
    Parameters:
    
    trainer - the trainer
  - getTrainer
```
public Trainer getTrainer()
```
    Returns the trainer of the net.
    
    Returns:
    
    the trainer
  - getMaxCycles
```
public int getMaxCycles()
```
    Returns:
    
    the maximum number of cycles (iterations) per turn.
  - setMaxCycles
```
public void setMaxCycles(int maxCycles)
```
    Parameters:
    
    maxCycles - the maximum number of cycles (iterations) per turn.
  - setZeroInputAfterFirstCycle
```
public void setZeroInputAfterFirstCycle(boolean makeZero)
```
    Define whether to reset the external input of the input neurons to zero after the first cycle when innervating the network. This is useful for recursive networks.
    
    Default: false, i.e. do not reset to zero.
  - getZeroInputAfterFirstCycle
```
public boolean getZeroInputAfterFirstCycle()
```
    Return whether the NeuralNet zeroes out the external input of the input neurons after the first evaluation cycle.
  - load
```
public static NeuralNet load(java.io.File file,
                             IOFilter filter)
                      throws java.io.IOException
```
    Loads a NeuralNet from the given file using the set file filter. If loading is successful, the given filter is attached to the net. Do not forget to configure the filter for the correct state of compression of the file to be loaded.
    
    Parameters:
    
    file - file to load the network from
    
    filter - the filter
    
    Returns:
    
    this neural network, null if problems
    
    Throws:
    
    java.io.IOException
  - save
```
public void save(java.io.File file)
          throws java.io.IOException
```
    Save a NeuralNet to the given file, in the Boone XNet format, using the IOFilter class.
    
    Throws:
    
    java.io.IOException
  - fromXML
```
public void fromXML(org.jdom2.Element element)
```
    Loads the network attributes from XML.
    
    Specified by:
    
    fromXML in interface Storable
    
    Parameters:
    
    element - the main net 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 net element
  - clone
```
public NeuralNet clone()
```
    Clone method for this NeuralNet. Returns a deep copy, i.e. duplicates all Neurons, Links, Trainers etc.
    
    Overrides:
    
    clone in class java.lang.Object
    
    Throws:
    
    CloneException - if cloning fails internally.
  - linkNet
```
public void linkNet(NeuralNet other)
```
    Links another net to this net (by appending at end).
    Links: just propagating neuron output over link without changing it.
    
    Parameters:
    
    other - the net that will be linked

Class NeuralNet

Field Summary

Constructor Summary

Method Summary

Methods inherited from class java.lang.Object

Field Detail

UNKNOWN

FEEDFORWARD

RECURRENT

topology

layers

neurons

inputNeurons

outputNeurons

links

trainer

maxCycles

zeroInputAfterFirstCycle

filter

props

Constructor Detail

NeuralNet

Method Detail

getTopology

setTopology

getFilter

setFilter

getNeuronCount

getNeuron

getNeuronIndex

getNeuronID

getInputNeuronCount

getInputNeuron

getOutputNeuronCount

getOutputNeuron

getOutputNeuronIndex

getHiddenNeuronCount

addNeuron

addLayer

replaceNeuron

removeNeuron

getLinkCount

getLink

getLinkID

addLink

replaceLink

removeLink

setInput

setInput

getOutput

randomize

innervate

toString

setTrainer

getTrainer

getMaxCycles

setMaxCycles

setZeroInputAfterFirstCycle

getZeroInputAfterFirstCycle

load

save

fromXML

toXML

clone

linkNet