AbstractNeuron Class Reference

Implementation helper for common types of Neurons. More...

#include <AbstractNeuron.h>

Inheritance diagram for AbstractNeuron:

[legend]Collaboration diagram for AbstractNeuron:

[legend]List of all members.

Public Member Functions
	AbstractNeuron (int label, bool hasBias=true)
	Creates a simple neuron with the given label.
virtual void	setBias (real bias)
	Sets the bias of the neuron.
virtual bool	hasBias () const
	Is the neuron allowed to have a bias?
virtual real	getBias () const
	The bias of the neuron, 0.0 if it's not allowed to have a bias.
virtual void	removeBias ()
	Sets bias to 0.0 and prevents the neuron from having a bias.
virtual void	setDesiredOutput (real desired)=0
	Sets the desired output of the neuron.
virtual void	connect (Neuron *from)
	Connects the given neuron to this one, i.e., the output of the supplied neuron will be given as input to this neuron.
void	connect (Neuron &from)
void	connect (Neuron &from, real w)
void	connectFrom (Neuron *from)
	TODO: rename the above...
virtual void	connect (Neuron *from, real weight)
	Connects the given neuron to this one, i.e., the output of the supplied neuron will be given as input to this neuron.
virtual void	calculateNewWeights (real learningRate, real momentum)
	Calculates the adjustment to incoming weights based on the gradient descent rule (backpropagation).
virtual void	update ()
	Updates the weights of incoming connections according to the values calculated using the gradient descent rule.
virtual const char *	getClassName () const
	Returns "AbstractNeuron".
virtual real	getWeight (Neuron *from) const
	Returns the weight of the link between the provided neuron and this neuron.
void	randomizeWeights ()
virtual	operator std::string () const
	Neuron::string().
Static Public Member Functions
real	getRandomWeight ()
Protected Member Functions
virtual void	_recacheOutput () const=0
	Neuron::_recacheOutput.
virtual void	_recacheError () const=0
	Neuron::_recacheError.
Protected Attributes
bool	_hasBias
	Is this neuron allowed to have a bias?
real	_bias
	If allowed to have a bias then the bias, otherwise 0.0.
real	_deltaBias
	The change in bias, calculated using the gradient descent rule.
Static Protected Attributes
const real	INIT_WEIGHT_MAX

---

Detailed Description

Implementation helper for common types of Neurons.

Has everything except activation function, which is defined in offsprings.

Note: rather than adding more virtual functions for activation function and derivation, we make virtual recaching functions. This should be more efficient in the performance-critical parts.

---

Constructor & Destructor Documentation

AbstractNeuron (  int  label, bool  hasBias = true )

Creates a simple neuron with the given label. Parameters: label  The label to be given to the neuron hasBias  true if the neuron is allowed to have a bias, false otherwise. Default is true See also: removeBias

---

Member Function Documentation

virtual void _recacheError (   )  const [protected, pure virtual]

Neuron::_recacheError. Implements Neuron. Implemented in SimpleNeuron.

virtual void _recacheOutput (   )  const [protected, pure virtual]

Neuron::_recacheOutput. Implements Neuron. Implemented in KohonenNeuron, and SimpleNeuron.

virtual void calculateNewWeights (  real  learningRate, real  momentum )  [virtual]

Calculates the adjustment to incoming weights based on the gradient descent rule (backpropagation). Doesn't actually update the weights, just sets the value. See also: update

virtual void connect (  Neuron *  from, real  weight )  [virtual]

Connects the given neuron to this one, i.e., the output of the supplied neuron will be given as input to this neuron. The weight of the link will be the one supplied. Parameters: from  The neuron whose output is to be taken as input weight  The weight of the connection

void connect (  Neuron &  from, real  w )  [inline]

Here is the call graph for this function:

void connect (  Neuron &  from  )  [inline]

Here is the call graph for this function:

virtual void connect (  Neuron *  from  )  [virtual]

Connects the given neuron to this one, i.e., the output of the supplied neuron will be given as input to this neuron. A random weight is provided to the link Parameters: from  The neuron whose output is to be taken is as input

void connectFrom (  Neuron *  from  )  [inline]

TODO: rename the above...

Here is the call graph for this function:

virtual real getBias (   )  const [virtual]

The bias of the neuron, 0.0 if it's not allowed to have a bias.

virtual const char* getClassName (   )  const [virtual]

Returns "AbstractNeuron". Implements Neuron. Reimplemented in RecurrentNeuron, and SimpleNeuron.

real getRandomWeight (   )  [static]

virtual real getWeight (  Neuron *  from  )  const [virtual]

Returns the weight of the link between the provided neuron and this neuron. Parameters: from  The neuron whose output is connected to this neuron's input Returns: The weight of the connection, 0.0 if no connection exists Reimplemented from Neuron.

virtual bool hasBias (   )  const [virtual]

Is the neuron allowed to have a bias?

virtual operator std::string (   )  const [virtual]

Neuron::string(). Reimplemented from Neuron.

void randomizeWeights (   )

Reimplemented in KohonenNeuron.

virtual void removeBias (   )  [virtual]

Sets bias to 0.0 and prevents the neuron from having a bias.

virtual void setBias (  real  bias  )  [virtual]

Sets the bias of the neuron. Parameters: bias  The bias to be given to the neuron Exceptions: Exception  if the neuron is not allowed to have a bias

virtual void setDesiredOutput (  real  desired  )  [pure virtual]

Sets the desired output of the neuron. Should be called only for output neurons, i.e., those whose output is not connected to anyone else. Setting the desired output at these neurons will form the basis of error backpropagation Parameters: desired  The desired output of this neuron Exceptions: Exception  if the neuron is not an output neuron Implemented in ParametrizedNeuron, and SimpleNeuron.

virtual void update (   )  [virtual]

Updates the weights of incoming connections according to the values calculated using the gradient descent rule. See also: calculateNewWeights Reimplemented in RecurrentNeuron.

---

Member Data Documentation

real _bias [protected]

If allowed to have a bias then the bias, otherwise 0.0. Reimplemented in RecurrentNeuron.

real _deltaBias [protected]

The change in bias, calculated using the gradient descent rule.

bool _hasBias [protected]

Is this neuron allowed to have a bias?

const real INIT_WEIGHT_MAX [static, protected]

---

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

• AbstractNeuron.h

---