ann Namespace Reference

Artificial Neural Networks (ANN) building blocks. More...

Classes
struct	Config
	The ANN library configuration. More...

Enumerations
enum	ActivationFunction {   ActivationFunction::Identity,   ActivationFunction::Logistic,   ActivationFunction::Tanh,   ActivationFunction::ReLU,   ActivationFunction::Neat,   ActivationFunction::ReExp,   ActivationFunction::LogisticEx }
	The types of supported activation functions. More...

Functions
void	setActivationFunction (ActivationFunction afn)
	Selects the activation function.
void	setGateActivationFunction (ActivationFunction afn)
	Selects the gate activation function (used with ANNs which include gates, for example LSTM)
float	activate (float x)
	Applies the selected activation function. More...
float	activateGate (float x)
	Applies the selected gate activation function. More...
float	afnIdentity (float x)
	Identity function.
float	afnLogistic (float x)
	Standard logistic function.
float	afnTanh (float x)
	tanh
float	afnReLU (float x)
	ReLU.
float	afnNeat (float x)
	Classic NEAT activation function.
float	afnReExp (float x)
	Experimental: ReExp. More...
float	afnLogisticEx (float x)
	Experimental: Yet another sigmoid function. More...
template<class T >
void	reset (std::vector< T > &v)
	Reset the values in a vector to 0.
void	reset (Matrix &matrix)
	Reset the values in a Matrix.
void	activateLayer (vector< float > &out)
	Apply the activation function over a set of values. More...
void	randomize (Matrix &w)
	Randomize the values in a Matrix. More...
float	roundWeight (float w)
	Ajust a value by rounding to Config::connection_resolution.
template<class T , class RND >
void	mutateValue (T &value, RND &rnd, T std_dev)
	Mutate a value. More...

Variables
EvaluateLayer	evaluateLayer = nullptr
	Evaluate a fully connected layer.

Detailed Description

Artificial Neural Networks (ANN) building blocks.

Enumeration Type Documentation

ActivationFunction

enum ann::ActivationFunction

strong

The types of supported activation functions.

Note

Consider adding new activation functions to cgp::FunctionId as well

Enumerator
Identity	Identity.
Logistic	Logistic.
Tanh	Hyperbolic tangent (tanh)
ReLU	ReLU.
Neat	NEAT activation function.
ReExp	Experimental.
LogisticEx	Experimental.

Function Documentation

activate()

float ann::activate ( float  x )

inline

Applies the selected activation function.

See also

setActivationFunction

activateGate()

float ann::activateGate ( float  x )

inline

Applies the selected gate activation function.

See also

setGateActivationFunction

activateLayer()

void ann::activateLayer ( vector< float > &  out )

inline

Apply the activation function over a set of values.

See also

ann::activate()

afnLogisticEx()

float ann::afnLogisticEx ( float  x )

inline

Experimental: Yet another sigmoid function.

afnReExp()

float ann::afnReExp ( float  x )

inline

Experimental: ReExp.

mutateValue()

template<class T , class RND >

void ann::mutateValue	(	T &	value,
		RND &	rnd,
		T	std_dev
	)

Mutate a value.

See also

Config::mutation_normal_distribution

Config::connection_range

roundWeight()

randomize()

void ann::randomize ( Matrix &  w )

inline

Randomize the values in a Matrix.

• The random values are in the [-Config::connection_range, Config::connection_range]
• The values are rounded using ann::roundWeight()
• If Config::sparse_weights is true, only a subset of values are non-zero (subject to the Config::weights_density value)