Description

This plugin aims to provide an easy and flexible artificial neural networks framework.

Elements

-NeuralNet
-BaseNeuralNet
-NeuralNetMethods
-ForEachConnectionHandle

Documentacion

Do not call functions with “_” prefix directly.

NeuralNet

Inherits:

BaseNeuralNet < Reference

Description

Ready to use net with sigmoidal activation function f(x)=x/sqrt(1+x^2).

BaseNeuralNet

Inherits:

Reference

Description

Semi “abstract” class. Base for all neural networks.

Properties

Array net_structure - Array of ints describing each layer size. setget will return or pass duplicate

Array data - Array of arrays of floats, value of each connection (weight). setget will return or pass duplicate

Array neuron_dat - Array of arrays of floats, last neuron value or neuron error value, depends on whether was .calculate() or .back_propagation() method called last, setget will return or pass duplicate

Methods

BaseNeuralNet new(Array net_structure) - Creates a new object sets self.net_structure to duplicate of net_structure. Sets value of each connection to 0.

void add_random(float min_val, float max_val) - Adds random value to each connection.

void multiply_random(float min_val, float max_val) - Multiplies each connection by random value.

Array calculate(Array input) - Runs net. Returns Array of floats of size of last self.net_structure cell value.

void back_propagation(Array error, float lerning_factor) - Propagates error and correct connections.

float _actiavte(float e) - “Virtual” function. Default return value is always zero. Override it with activation function of your choice.

float _derivative(float x) - “Virtual” function. Derivative of activation function in point x. No need to overwrite it if you don't plan to use back propagation.

void for_each_connection(ForEachConnectionHandle handle) - Calls handle._for_each_connection(int x, int y, int z, BaseNeuralNet net) for each connection where x and y (x is closer to or is in the output layer) are neurons that it is connecting. Z is the first self.data index and index is second. In another words connection is data[z][index]. Index is calculated from this formula i = x*(net_structure[z]+1)+y. Net is self.

NeuralNetMethods

Inherits:

Node

Description

Singleton container for methods that would cause cyclic dependency errors if placed as member functions.

Methods

String print_net(BaseNeuralNet net) - converts net to JSON string.

BaseNeuralNet parse_net(String json) - converts String back to BaseNeuralNet.

ForEachConnectionHandle

Inherits:

Reference

Description

“Abstract” class. Provides way to change connections (weights) in a custom way.

Methods

void _for_each_connection(int x, int y, int z, int index, BaseNeuralNet net)

How to