#ifndef NEURON_H #define NEURON_H //The Neuron class Neuron { //what does the Neuron do?? //Its Principle is not that difficult: //It gets maxI Inputs(I will call them xI) and multiplies each of them with a wheight (wI) //so first Input*first wheight and so on.... //After that it just sums all of this products up //These sum is put in the activations function //in this Neural Network the activation function is the sigmoid function, but there are many of them //The result of this function is called activation of this Neuron //sigmoid function: //sigmoid function gives an result between 0 and 1: //sigmoid(-infinity)=0 //sigmoid(infinity)=1 //sigmoid(0)=0.5 //Example like here: //https://www.sololearn.com/post/359703/?ref=app //Lets take some example: //The Inputs could be coordinates in an 2-Dimensional room (2 Inputs:x=3,y=2) //The wheights could be 1 and -1 //The output could be the suggested colour of the point: 0 for blue and 1 for red //So what the Neuron is doing is just calculating: sum=x*1+y*-1=x-y and putting this into sigmoid function, so if x-y>0 then this Neuron suggests the point to be more like red, otherwise more like blue.... //So it ould be like this: //rrrrrr //brrrrr //bbrrrr //bbbrrr //bbbbrr //bbbbbr //bbbbbb //But what if P(3,2) is blue??//Thie means that the wheights are probably wrong //And this is what the learn() does: It tries to improve the wheights! //For this it needs altleast a pair of inputs and the mistake it has made(correctOutput-output)... //Then it changes every wheight a little bit to improve the output: //(wheight I)+="Learnrate"*mistake*(Input I) for every wheight //if the mistake is positive the wheight and through this the resulted output will be incresed, otherwise decreased, so after the learning proccess the data fits better... //"Learnrate" steers how much the wheights are changed, too much changing is bad //(imagine the Neuron is perfect trained, only one point does not fit and destroys all well-trained wheights) //Why do we multiply with mistake and Input I?? //We multiply with mistake for the same reason as there is a Learnrate, the better the output already is the more we want to leave it as it is //We multiply with Input I, because the higher Input I is the higher is the effect of changing the wheight, for example we dont want to change a wheight with input 0, because this would not improve the output private: //These are the wheights of the Neuron double *wI; //This is the numebr of Inputs int maxI; //This is the activation double activation; public: //The constructor and destructor are just empty Neuron(); virtual ~Neuron(); //just initializes the maxI and allocates place for the wheights void initialize(int n); //calculates void calc(double* xI); void calc(int* xI); //improve the weights void learn(double Fehler, double *xI); void learn(double Fehler, int *xI); //stupid getfunctions which return values double getactivation(); double* getwI(); protected: }; #endif // NEURON_H #include "stdlib.h" #include #include Neuron::Neuron() {} Neuron::~Neuron() {} void Neuron::initialize(int n) //initializes the Neuron { maxI=n; //with n inputs wI=new double[n+1]; //and n+1 weights (one for every input abd one bias) for(int i=0;i0) //activation=sum; //sets the activation to the sum } void Neuron::calc(int *xI) //the same as up just with integers { double sum=wI[maxI]; //here the bias is ad directly for(int i=0;i0) //activation=sum; } double Neuron::getactivation() //returns the activation from the Neuron { return activation; } void Neuron::learn(double mistakes,double *xI) //learns with the mistakes and the inputs { for(int i=0;i Layer::Layer() { } Layer::~Layer() { //dtor } void Layer::initialize(int nN, int nxI) //initializes the Layer and their Neurons { nneurons=nN; //saves the number of the Neurons neurons=new Neuron[nN]; //constructs the Neurons for(int i=0;i Network::Network(int nL, int nN, int nXi) { nNeurons=nN; nLayer=nL; layers= new Layer[nL]; for(int i=0; i0;i--) { layers[i].learn(Fehler,activations[i-1]); for(int j=0;j #include #include using namespace std; int main() { //Here you can change the Number of Neurons: int nN=3; double *inputs= new double[nN]; srand(time(0)); double *n=new double[nN]; double *outputdesired= new double[nN]; //for(int i=0;i

Neural network