The Neural Network Class. More...

Inheritance diagram for NeuralNet:

Public Member Functions
	NeuralNet ()
	Construct an empty neural network.
	~NeuralNet ()
	Destructor.
void	addCluster (Cluster *c, bool isInput=false, bool isOutput=false)
	Add a Cluster into the neural network If isInput is true then the Cluster will be considered as an Input Cluster of this network If isOutput is true then the Cluster will be considered as an Output Cluster of this network.
void	addInputCluster (Cluster *c)
	Add a Cluster and mark it as Input Behave exactly the same of addCluster( c, true, false )
void	addLinker (Linker *l)
	Add Linker.
void	addOutputCluster (Cluster *c)
	Add a Cluster and mark it as Output Behave exactly the same of addCluster( c, false, true )
template<class PointerTo >
PointerTo	byName (QString aName, PointerTo &aPointer)
	Search into the net for the presence of an Updatable with name aName; on success set the pointer aPointer and return it, otherwise it set aPointer to zero and return zero.
ClusterList	clusters () const
	Returns the vector of Clusters contained.
virtual void	configure (ConfigurationParameters &params, QString prefix)
	Configures the object using a ConfigurationParameters object.
bool	find (const Cluster *) const
	Return true if the Cluster is in this net.
bool	find (const Updatable *) const
	Return true if the Updatable object is in this net.
bool	find (const Linker *) const
	Return true if the Linker is in this net.
Updatable *	getByName (QString)
	Return the Updatable with the name specified Returns NULL-pointer if there's no updatable object whit the name specified
ClusterList	hiddenClusters () const
	Returns the vector of Hidden Clusters contained (i.e.
ClusterList	inputClusters () const
	Returns the vector of Input Clusters contained.
bool	isIsolated (Cluster *c) const
	Return true if there isn't any Linker connected with Cluster c.
LinkerList	linkers () const
	Returns the array of Linkers contained.
LinkerList	linkers (Cluster *c, bool out=false) const
	If out is true, return the Linkers outgoing from Cluster c, otherwise return incoming Linkers.
void	markAsInput (Cluster *c)
	Mark a Cluster as an Input Cluster of this network.
void	markAsOutput (Cluster *c)
	Mark a Cluster as an Output Cluster of this network.
QString	name ()
	return the name of the NeuralNet
UpdatableList	order () const
	Return the order.
ClusterList	outputClusters () const
	Returns the vector of Output Clusters contained.
void	randomize (double min, double max)
	This randomize the free parameters of the all elements of the neural net This method call randomize method of every Cluster and Linker inserted.
bool	removeCluster (Cluster *c)
	Remove a Cluster from the network.
bool	removeLinker (Linker *)
	Remove Linker.
virtual void	save (ConfigurationParameters &params, QString prefix)
	Save the actual status of parameters into the ConfigurationParameters object passed.
void	setName (QString name)
	Set the name of the NeuralNet It's useful when loading from ConfigurationParameters more than NeuralNet or from LearningAlgorithms subclasses.
void	setOrder (const UpdatableList &)
	Set the order.
void	setOrder (Updatable *updatables[], unsigned int dim)
	Set the order.
void	step ()
	Step.
void	unmark (Cluster *c)
	Eliminate the marks from Cluster passed.
void	unmarkAll ()
	Eliminate the marks from all Cluster present in this networks.
Static Public Member Functions
static void	describe (QString type)
	Add to Factory::typeDescriptions() the descriptions of all parameters and subgroups.
Protected Types
typedef QMap< QString, Cluster * >	ClustersMap
typedef QMap< QString, Linker * >	LinkersMap
typedef QMap< Cluster *, LinkerList >	LinkVecMap
Protected Attributes
ClustersMap	clsMap
	map name -> Cluster*
ClusterList	clustersv
	Clusters.
unsigned int	dimUps
ClusterList	hidclusters
	unmarked Clusters
ClusterList	inclusters
	Input Clusters.
LinkVecMap	inLinks
	mappa dei linkers entranti (cluster -> vettore linkers entranti)
LinkerList	linkersv
	Linkers.
LinkersMap	lksMap
	map name -> Linker*
QString	namev
	name of this NeuralNet
ClusterList	outclusters
	Output Clusters.
LinkVecMap	outLinks
	map of outgoing linkers (cluster -> vettore linkers uscenti)
UpdatableList	ups
	Array of Updateables ordered as specified.

Detailed Description

The Neural Network Class.

The NeuralNet class can seen as a simple container of Clusters and Linkers
The relationship among Clusters and Linkers must be specified by cluster and linker constructors;

When the NeuralNet is configured from a file, it is configured in the following way:

the parameters for configuring a NeuralNet are: inputClusters, outputClusters, spreadOrder, clustersList, linkersList; all parameters consist in a list of the group name where is present the corresponding object configuration
the only mandatory parameter is spreadOrder
inputClusters is the list of all Clusters considered the input layer of the NeuralNet
outputClusters is the list of all Clusters considered the output layer of the NeuralNet
clustersList and LinkersList are the full list of all Clusters and Linkers to put into the network; in some cases, these parameters are redundant, because if a Cluster or a Linker is specified into spreadOrder then it will be automatically put into to the NeuralNet
all sub-groups of the group containing the parameters of NeuralNet are considered object to be created and to be put into the NeuralNet

Some examples of configuration files using the INI format. The first example shows how to create a NeuralNet with two Cluster as input and one Cluster as output connected by two linkers. This example uses all parameters even if some are redundant

 inputClusters = input1 input2
 outputClusters = output1
 spreadOrder = input1 input2 linker1 linker2 output1
 clustersList = input1 input2 output1
 linkersList = linker1 linker2
 
 [input1]
 type = FakeCluster
 
 [input2]
 type = SimpleCluster
 [input2/OutFunction]
 type = StepFunction
 min = -1.0
 max = +1.0
 threshold = 0.5
 
 [output1]
 type = BiasedCluster
 [output1/OutFunction]
 type = SigmoidFunction
 lambda = 0.5
 
 [linker1]
 type = DotLinker
 from = input1
 to = output1
 
 [linker2]
 type = DotLinker
 from = input2
 to = output1

The same NeuralNet can be configured without specifing clustersList and linkersList with the following file INI:

 inputClusters = input1 input2
 outputClusters = output1
 spreadOrder = input1 input2 linker1 linker2 output1
 
 [input1]
 type = FakeCluster
 
 [input2]
 type = SimpleCluster
 [input2/OutFunction]
 type = StepFunction
 min = -1.0
 max = +1.0
 threshold = 0.5
 
 [output1]
 type = BiasedCluster
 [output1/OutFunction]
 type = SigmoidFunction
 lambda = 0.5
 
 [linker1]
 type = DotLinker
 from = input1
 to = output1
 
 [linker2]
 type = DotLinker
 from = input2
 to = output1

Let's suppose that you want to change the NeuralNet of the example above removing input1 as input Cluster of the NeuralNet then the two following configurations results in the same NeuralNet:

 inputClusters = input2
 outputClusters = output1
 spreadOrder = input1 input2 linker1 linker2 output1
 
 ...

 inputClusters = input2
 outputClusters = output1
 spreadOrder = input2 linker1 linker2 output1
 
 ...

Because even if in the last example the input1 is not specified in any of the NeuralNet parameters, it will be created because [input1] is a subgroup of the main group where the NeuralNet parameters are present

To understand better this situation, let's put all NeuralNet parameters into a group called [Net1], the above example will appear in this way:

 [Net1]
 inputClusters = input2
 outputClusters = output1
 spreadOrder = input2 linker1 linker2 output1
 
 [input1]
 type = FakeCluster
 
 [input2]
 type = SimpleCluster
 [input2/OutFunction]
 type = StepFunction
 min = -1.0
 max = +1.0
 threshold = 0.5
 
 [output1]
 type = BiasedCluster
 [output1/OutFunction]
 type = SigmoidFunction
 lambda = 0.5
 
 [linker1]
 type = DotLinker
 from = input1
 to = output1
 
 [linker2]
 type = DotLinker
 from = input2
 to = output1

Now the input1 will be not automatically added to the NeuralNet because the group [input1] is not anymore a subgroup of the group ([Net1]) where the NeuralNet parameters are present. In this case, if you want that also [input1] will be put into the NeuralNet you have two options:

change the group of [input1] to [Net1/input1]; but this also means that the corresponding Cluster will be named "Net1/input1" instead of "input1"

add a clustersList parameter where you declare the complete list of Clusters including input1; in this way the name of the Cluster will remain "input1" The two alternatives appear in the following way:

 [Net1]
 inputClusters = input2
 outputClusters = output1
 spreadOrder = input2 linker1 linker2 output1
 
 [Net1/input1]
 type = FakeCluster
 ...

 [Net1]
 inputClusters = input2
 outputClusters = output1
 spreadOrder = input2 linker1 linker2 output1
 clustersList = input1 input2 output1
 
 [input1]
 type = FakeCluster
 ...

Definition at line 209 of file neuralnet.h.

Constructor & Destructor Documentation

NeuralNet ( )

Construct an empty neural network.

Definition at line 24 of file neuralnet.cpp.

~NeuralNet ( )

Destructor.

Definition at line 28 of file neuralnet.cpp.

Member Function Documentation

void addCluster	(	Cluster *	c,
		bool	isInput = `false`,
		bool	isOutput = `false`
	)

Add a Cluster into the neural network
If isInput is true then the Cluster will be considered as an Input Cluster of this network
If isOutput is true then the Cluster will be considered as an Output Cluster of this network.

Definition at line 31 of file neuralnet.cpp.

References NeuralNet::clsMap, NeuralNet::clustersv, NeuralNet::find(), NeuralNet::hidclusters, NeuralNet::inclusters, Updatable::name(), and NeuralNet::outclusters.

Referenced by NeuralNet::configure().

void addInputCluster ( Cluster * c ) [inline]

Add a Cluster and mark it as Input
Behave exactly the same of addCluster( c, true, false )

Definition at line 232 of file neuralnet.h.

void addLinker ( Linker * l )

Add Linker.

Definition at line 172 of file neuralnet.cpp.

References NeuralNet::find(), Linker::from(), NeuralNet::inLinks, NeuralNet::linkersv, NeuralNet::lksMap, Updatable::name(), NeuralNet::outLinks, and Linker::to().

Referenced by NeuralNet::configure().

void addOutputCluster ( Cluster * c ) [inline]

Add a Cluster and mark it as Output
Behave exactly the same of addCluster( c, false, true )

Definition at line 237 of file neuralnet.h.

PointerTo byName	(	QString	aName,
		PointerTo &	aPointer
	)		`[inline]`

Search into the net for the presence of an Updatable with name aName; on success set the pointer aPointer and return it, otherwise it set aPointer to zero and return zero.

This allow to use it both into an if-statement and an assignment:

  BiasedCluster* bias1;
  BiasedCluster* bias2;
  if ( byName("aName", bias1 ) ) {
      //--- ok, there is a BiasedCluster with name "aName"
      //--- now bias1 points to the BiasedCluster with name "aName"
  } else {
      //--- error, there is no BiasedCluster with that name
      //--- now bias1 is NULL
  }
  //--- you can also use it for assignment:
  bias2 = byName("aName", bias1);

Definition at line 308 of file neuralnet.h.

ClusterList clusters ( ) const

Returns the vector of Clusters contained.

Definition at line 156 of file neuralnet.cpp.

References NeuralNet::clustersv.

void configure	(	ConfigurationParameters &	params,
		QString	prefix
	)		`[virtual]`

Configures the object using a ConfigurationParameters object.

Parameters:

params	the configuration parameters object with parameters to use
prefix	the prefix to use to access the object configuration parameters. This is guaranteed to end with the separator character when called by the factory, so you don't need to add one

Implements ParameterSettableWithConfigureFunction.

Definition at line 320 of file neuralnet.cpp.

References NeuralNet::addCluster(), NeuralNet::addLinker(), ConfigurationParameters::getGroupsList(), ConfigurationParameters::getObjectFromGroup(), ConfigurationParameters::getValue(), ConfigurationParameters::GroupSeparator(), NeuralNet::markAsInput(), NeuralNet::markAsOutput(), NeuralNet::setName(), ConfigurationParameters::startRememberingGroupObjectAssociations(), ConfigurationParameters::stopRememberingGroupObjectAssociations(), and NeuralNet::ups.

void describe ( QString type ) [static]

Add to Factory::typeDescriptions() the descriptions of all parameters and subgroups.

Reimplemented from ParameterSettable.

Definition at line 453 of file neuralnet.cpp.

References ParameterSettable::addTypeDescription(), ParameterSettable::Descriptor::describeObject(), ParameterSettable::ObjectDescriptor::help(), ParameterSettable::IsList, ParameterSettable::IsMandatory, ParameterSettable::ObjectDescriptor::props(), and ParameterSettable::ObjectDescriptor::type().

bool find ( const Cluster * cl ) const

Return true if the Cluster is in this net.

Definition at line 305 of file neuralnet.cpp.

References NeuralNet::clustersv.

Referenced by NeuralNet::addCluster(), NeuralNet::addLinker(), NeuralNet::markAsInput(), NeuralNet::markAsOutput(), NeuralNet::removeCluster(), NeuralNet::removeLinker(), NeuralNet::setOrder(), and NeuralNet::unmark().

bool find ( const Updatable * u ) const

Return true if the Updatable object is in this net.

Definition at line 313 of file neuralnet.cpp.

References NeuralNet::clustersv, and NeuralNet::linkersv.

bool find ( const Linker * l ) const

Return true if the Linker is in this net.

Definition at line 309 of file neuralnet.cpp.

References NeuralNet::linkersv.

Updatable * getByName ( QString name )

Return the Updatable with the name specified
Returns NULL-pointer if there's no updatable object whit the name specified

Warning:: return the first that finds. If you have named different Updatables with same name there no way to retrieve all of them with this methods... call them with unique name ;-)

Definition at line 292 of file neuralnet.cpp.

References NeuralNet::clsMap, NeuralNet::lksMap, and NeuralNet::name().

ClusterList hiddenClusters ( ) const

Returns the vector of Hidden Clusters contained (i.e.

UnMarked Clusters)

Definition at line 168 of file neuralnet.cpp.

References NeuralNet::hidclusters.

ClusterList inputClusters ( ) const

Returns the vector of Input Clusters contained.

Definition at line 160 of file neuralnet.cpp.

References NeuralNet::inclusters.

Referenced by BackPropagationAlgo::calculateMSE(), and BackPropagationAlgo::learn().