nn_base.h header file

This file declares all the elements needed to create and use generic (acyclic, non-recurrent) neural networks.

Classes:

class Weight: public Link

a weight in a neural network: it can calculate its output and the partial derivative of the network with respect to itself (using BP)

class Neuron: public Node

a neuron: holds the couple (activation function, derivative) and implements BP

class InputTerminal: public Neuron

an input terminal is a neuron without incoming links: it has a value of its own, and is used to feed input into neurla networks

class NeuralNet: public Graph

this holds neurons and weights, with methods to set the input and retrieve the output

class NeuronInserter

helper class to load neurons from a stream: inserts them into a set and a vector

class ITInserter

helper class to load input terminals from a stream: inserts them into a set and a vector

Functions:

• Function Streaming

  registerFun()

  associates a function with a name, so as to assign to each function a representation invariant between runs of the program

  getFun()

  returns the function pointer associated with the given name

  getName()

  returns the name associated with the given function pointer

  registerDefault()

  performs the predefined associations between to functions below and their names

• Standard Activation Functions

  id()

  identity function (used for input terminals)

  one()

  costant 1 function (derivative of the above)

  sigmoid()

  sigmoid function (asymmetric)

  dsigmoid()

  derivative of the above