nnfw::Cluster

Cluster Class Reference

Define the common interface among Clusters. More...

Inheritance diagram for Cluster:

[legend]Collaboration diagram for Cluster:


Constructors
	Cluster (u_int numNeurons, const char *name="unnamed")
	Cluster (PropertySettings &prop)
virtual	~Cluster ()
Methods affecting whole Cluster
u_int	numNeurons () const
bool	needReset ()
void	setAccumulate (bool mode)
bool	isAccumulate () const
virtual void	randomize (Real min, Real max)=0
Variant	numNeuronsP ()
Variant	accumP ()
bool	setAccumP (const Variant &b)
Operations on Input's vector
virtual void	setInput (u_int neuron, Real value)
virtual void	setInputs (const RealVec &inputs)
virtual void	setAllInputs (Real value)
virtual void	resetInputs ()
virtual Real	getInput (u_int neuron) const
RealVec &	inputs ()
const RealVec &	inputs () const
Variant	inputsP ()
bool	setInputsP (const Variant &v)
Operations on Output's vector
virtual void	setOutput (u_int neuron, Real value)
virtual void	setOutputs (const RealVec &outputs)
virtual Real	getOutput (u_int neuron) const
RealVec &	outputs ()
const RealVec &	outputs () const
Variant	outputsP ()
bool	setOutputsP (const Variant &v)
Operations on OutputFunction
void	setFunction (const OutputFunction &up)
OutputFunction *const	getFunction () const
Variant	getFunctionP ()
bool	setFunction (const Variant &outf)
virtual Cluster *	clone () const
Protected Member Functions
void	setNeedReset (bool b)

Detailed Description

Motivation: The Cluster class define the common interface amog Cluster. The subclasses may extends this interface for specific purpose (ex. SimpleCluster), but the BaseNeuralNet, the Linker and other classes depends only by this class. This abstraction allow to isolate the specific implementation of various classes

Description

The Cluster class represent an abstract group of neurons. There isn't a neuron class. The Cluster class represent a group of neurons as two arrays: inputs and outputs. The inputs array represent the inputs of the neurons 'contained' in the cluster, and the outputs of this neurons are calculated by appling the function provided by OutputFunction.
The number of neuron returned by size() method is also the dimension of inputs and outputs arrays
You can sets one subclasses of OutputFunction by setUpdater methods. If you don't specify an index when set a OutputFunction then this OutputFunction will be used to update the output of all neurons. Otherwise, you can specifiy different OutputFunction for different neuron.

 // create a SimpleCluster, a specialized subclass of Cluster
 SimpleCluster* simple = new SimpleCluster( 10 ); // this cluster contains 10 neurons
 // set the SigmoidUpdater for all neurons
 simple->setUpdater( SigmoidUpdater( 1.0 ) );

Warnings

For whose want to implement a subclass of Cluster: This class allocate the memory for inputs, outputs and updaters array. So, a subclass have to implements only the update method. The getInputs and getOutputs returns a valid array of internal data, and not simply a copy of the internal data. Look at the following code:

 RealVec& in = cluster->inputs();
 in[2] = 3.0;   // This statement will be changes the inputs of third neuron.
 // the above statements must be equivalent with the following
 cluster->setInput( 2, 3.0 );

The reasons behind this kind of behaviour its the efficiency!! When another class must do heavy calculation on all inputs of a Cluster (as MatrixLinker do), then its more efficient that it takes the array returned by inputs (or outputs) and works over them.

Name	Type [isVector]	Access mode	Description	Class
Properties
typename	string	read-only	Class's type	Propertized
name	string	read/write	name of the object	Updatable
accumulate	boolean	read/write	if inputs are accumulated	this
inputs	RealVec	read/write	neuron's input	this
outfunction	OutputFunction	read/write	neuron's output function	this
outputs	RealVec	read/write	neuron's output	this
numNeurons	unsigned int	read-only	number of neurons	this

Constructor & Destructor Documentation

Cluster	(	u_int	numNeurons,
		const char *	name = `"unnamed"`
	)

Construct a Cluster

Cluster ( PropertySettings & prop )

Construct a Cluster with PropertySettings

virtual ~Cluster ( ) [virtual]

Destructor

Member Function Documentation

u_int numNeurons ( ) const [inline]

Return the number of neurons (the length of input and output arrays)

bool needReset ( ) [inline]

Return true if inputs needs a reset

void setAccumulate ( bool mode ) [inline]

Enable/Disable accumulation mode
If accumulation is enabled (true) then linkers attached to this Cluster will never resetInput and accumulates data, otherwise the inputs will be resetted at each step of neural network

bool isAccumulate ( ) const [inline]

return true if the Cluster will accumulates inputs

virtual void randomize	(	Real	min,
		Real	max
	)			`[pure virtual]`

Randomize the parameters of the Cluster
The parameters randomized by this method will be specified by sub-classes

Implemented in BiasedCluster, DDECluster, FakeCluster, and SimpleCluster.

Variant numNeuronsP ( ) [inline]

Read Access to property 'size'

Variant accumP ( ) [inline]

Read Access to property 'accumulate'

bool setAccumP ( const Variant & b ) [inline]

Write Access to property 'accumulate'

virtual void setInput	(	u_int	neuron,
		Real	value
	)			`[virtual]`

Set the input of neuron Details...

virtual void setInputs ( const RealVec & inputs ) [virtual]

Set the inputs from the vector given

virtual void setAllInputs ( Real value ) [virtual]

Set all the inputs with the same value Details...

virtual void resetInputs ( ) [virtual]

Reset the inputs of this cluster, typically this means that the inputs will be set to zero. Details...

virtual Real getInput ( u_int neuron ) const [virtual]

Get the input of neuron

RealVec& inputs ( ) [inline]

Get the array of inputs
Return the array of inputs, not a copy of inputs; Then you can change inputs by the pointer returned !!!

const RealVec& inputs ( ) const [inline]

const version of inputs()

Variant inputsP ( ) [inline]

For property 'inputs'

bool setInputsP ( const Variant & v ) [inline]

setting the property 'inputs'

virtual void setOutput	(	u_int	neuron,
		Real	value
	)			`[virtual]`

Force the output of the neuron at value specified

virtual void setOutputs ( const RealVec & outputs ) [virtual]

Set the outputs from the vector given

virtual Real getOutput ( u_int neuron ) const [virtual]

Get the output of neuron

RealVec& outputs ( ) [inline]

Get the array of outputs
Return the array of outputs, not a copy of outputs; Then you can change outputs by the pointer returned !!!

const RealVec& outputs ( ) const [inline]

const version of outputs()

Variant outputsP ( ) [inline]

For property 'outputs'

bool setOutputsP ( const Variant & v ) [inline]

setting the property 'outputs'

void setFunction ( const OutputFunction & up )

Set the output function for all neurons contained
This method create an internal copy of the OutputFunction passed

Warning:: This function delete the previous updater class registered !!!

OutputFunction* const getFunction ( ) const [inline]

Get the Output function

Variant getFunctionP ( ) [inline]

read property 'outfunction'

bool setFunction ( const Variant & outf ) [inline]

sets the property 'outfunction'

virtual Cluster* clone ( ) const [virtual]

Clone this Cluster

Reimplemented from Propertized.

Reimplemented in BiasedCluster, DDECluster, FakeCluster, and SimpleCluster.

void setNeedReset ( bool b ) [inline, protected]

Set the state of 'needReset'
Used by subclassed into update implementation

The documentation for this class was generated from the following file:

include/cluster.h

Contact: emmegian (at) yahoo (dot) it

Documentation generated by

Cluster Class Reference

Constructors

Methods affecting whole Cluster

Operations on Input's vector

Operations on Output's vector

Operations on OutputFunction

Protected Member Functions

Detailed Description

Constructor & Destructor Documentation

Member Function Documentation