Net is the main class used to set up a neural network used for closed-loop Deep Learning. More...

#include <Net.h>

Public Member Functions
	Net (const int _nLayers, const int *const _nNeurons, const int _nInputs, const int _subject, const string _trial)
	Constructor: The neural network that performs the learning. More...

	~Net ()
	Destructor De-allocated any memory.

void	initNetwork (Neuron::weightInitMethod _wim, Neuron::biasInitMethod _bim, Neuron::actMethod _am)
	Dictates the initialisation of the weights and biases and determines the activation function of the neurons. More...

void	setLearningRate (double _w_learningRate, double _b_learningRate)
	Sets the learning rate. More...

void	setInputs (const double *_inputs, const double scale=1.0, const unsigned int offset=0, const int n=-1)
	Sets the inputs to the network in each iteration of learning, needs to be placed in an infinite loop. More...

void	propInputs ()
	It propagates the inputs forward through the network.

void	propErrorBackward ()
	Propagates the error backward.

void	setError (double _leadError)
	Sets the error at the output layer to be propagated backward. More...

double	getGradient (Layer::whichGradient _whichGradient)
	It provides a measure of how the magnitude of the error changes through the layers to alarm for vanishing or exploding gradients. More...

void	updateWeights ()
	Requests that all layers perform one iteration of learning.

Layer *	getLayer (int _layerIndex)
	Allows Net to access each layer. More...

double	getOutput (int _neuronIndex)
	Allows the user to access the activation output of a specific neuron in the output layer only. More...

double	getSumOutput (int _neuronIndex)
	Allows the user to access the weighted sum output of a specific neuron in output layer only. More...

int	getnLayers ()
	Informs on the total number of hidden layers (excluding the input layer) More...

int	getnInputs ()
	Informs on the total number of inputs to the network. More...

double	getWeightDistance ()
	Allows for monitoring the overall weight change of the network. More...

double	getLayerWeightDistance (int _layerIndex)
	Allows for monitoring the weight change in a specific layer of the network. More...

double	getWeights (int _layerIndex, int _neuronIndex, int _weightIndex)
	Grants access to a specific weight in the network. More...

int	getnNeurons ()
	Informs on the total number of neurons in the network. More...

void	saveWeights ()
	Saves the temporal changes of all weights in all neurons into files.

void	snapWeights (string prefix, string _trial, int _subject)
	Snaps the final distribution of all weights in a specific layer, this is overwritten every time the function is called.

void	snapWeightsMatrixFormat (string prefix)

void	printNetwork ()
	Prints on the console a full tree of the network with the values of all weights and outputs for all neurons.

Detailed Description

Net is the main class used to set up a neural network used for closed-loop Deep Learning.

It initialises all the layers and the neurons internally.

GNU GENERAL PUBLIC LICENSE

Constructor & Destructor Documentation

◆ Net()

Net::Net	(	const int	_nLayers,
		const int *const	_nNeurons,
		const int	_nInputs,
		const int	_subject,
		const string	_trial
	)

Constructor: The neural network that performs the learning.

Parameters

_nLayers	Total number of hidden layers, excluding the input layer
_nNeurons	A pointer to an int array with number of neurons for all layers need to have the length of _nLayers.
_nInputs	Number of Inputs to the network

Member Function Documentation

◆ getGradient()

double Net::getGradient ( Layer::whichGradient _whichGradient )

It provides a measure of how the magnitude of the error changes through the layers to alarm for vanishing or exploding gradients.

Parameters

_whichError	choose what error to monitor, for more information see Neuron::whichError
_whichGradient	choose what gradient of the chosen error to monitor, for more information see Layer::whichGradient

Returns: Returns the ratio of the chosen gradient in the last layer to the the first layer

◆ getLayer()

Layer* Net::getLayer ( int _layerIndex )

Allows Net to access each layer.

Parameters

_layerIndex the index of the chosen layer

Returns: A pointer to the chosen Layer

◆ getLayerWeightDistance()

double Net::getLayerWeightDistance ( int _layerIndex )

Allows for monitoring the weight change in a specific layer of the network.

Parameters

_layerIndex The index of the chosen layer

Returns: returns the Euclidean wight distance of neurons in the chosen layer from their initial value

◆ getnInputs()

int Net::getnInputs ( )

Informs on the total number of inputs to the network.

Returns: Total number of inputs

◆ getnLayers()

int Net::getnLayers ( )

Informs on the total number of hidden layers (excluding the input layer)

Returns: Total number of hidden layers in the network

◆ getnNeurons()

int Net::getnNeurons ( )

Informs on the total number of neurons in the network.

Returns: The total number of neurons

◆ getOutput()

double Net::getOutput ( int _neuronIndex )

Allows the user to access the activation output of a specific neuron in the output layer only.

Parameters

_neuronIndex The index of the chosen neuron

Returns: The value at the output of the chosen neuron

◆ getSumOutput()

double Net::getSumOutput ( int _neuronIndex )

Allows the user to access the weighted sum output of a specific neuron in output layer only.

Parameters

_neuronIndex The index of the chosen neuron

Returns: The value at the sum output of the chosen neuron

◆ getWeightDistance()

double Net::getWeightDistance ( )

Allows for monitoring the overall weight change of the network.

Returns: returns the Euclidean wight distance of all neurons in the network from their initial value

◆ getWeights()

double Net::getWeights	(	int	_layerIndex,
		int	_neuronIndex,
		int	_weightIndex
	)

Grants access to a specific weight in the network.

Parameters

_layerIndex	Index of the layer that contains the chosen weight
_neuronIndex	Index of the neuron in the chosen layer that contains the chosen weight
_weightIndex	Index of the input to which the chosen weight is assigned

Returns: returns the value of the chosen weight

◆ initNetwork()

void Net::initNetwork	(	Neuron::weightInitMethod	_wim,
		Neuron::biasInitMethod	_bim,
		Neuron::actMethod	_am
	)

Dictates the initialisation of the weights and biases and determines the activation function of the neurons.

Parameters

_wim	weights initialisation method, see Neuron::weightInitMethod for different options
_bim	biases initialisation method, see Neuron::biasInitMethod for different options
_am	activation method, see Neuron::actMethod for different options

◆ setError()

void Net::setError ( double _leadError )

Sets the error at the output layer to be propagated backward.

Parameters

_leadError The closed-loop error for learning

◆ setInputs()

void Net::setInputs	(	const double *	_inputs,
		const double	scale = `1.0`,
		const unsigned int	offset = `0`,
		const int	n = `-1`
	)

Sets the inputs to the network in each iteration of learning, needs to be placed in an infinite loop.

Parameters

_inputs A pointer to the array of inputs

◆ setLearningRate()

void Net::setLearningRate	(	double	_w_learningRate,
		double	_b_learningRate
	)

Sets the learning rate.

Parameters

_learningRate Sets the learning rate for all layers and neurons.

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

Net.h

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ Net()

Member Function Documentation

◆ getGradient()

◆ getLayer()

◆ getLayerWeightDistance()

◆ getnInputs()

◆ getnLayers()

◆ getnNeurons()

◆ getOutput()

◆ getSumOutput()

◆ getWeightDistance()

◆ getWeights()

◆ initNetwork()

◆ setError()

◆ setInputs()

◆ setLearningRate()