InteligentAdaptativeFilter

Adaptative Filter for Any Objetive This is an adaptative filter, or a linear predictive model with machine learning, in the most easy way #We need to -> Define the class -> Give their data -> Train -> And Run

The magic are:

Instrucctions

• What number of elements do you want? # define BackFilter 10
• What will your Learning parameter? # define LearningParameter 0.3

The class that makes the magic

class FiltroInteligente{

//We have three variables with all the information
float Vin[BackFilter]; //Input values
float Kca[BackFilter]; //K parameters
float Pro[BackFilter]; //=inputs * K parameters
Method to clean the vectors
void ClearData(){
	for(int i=0; i<BackFilter;i++)	{
		Vin[i]=0;
		Pro[i]=0;
	}
}
Limpieza del vector de datos
void ClearProd(){
	for(int i=0; i<BackFilter;i++)	{
		Pro[i]=0;
	}
}
//Funcion para predecir los datos
public: 
float Result=0;
float Wantted=0;
float ErrorN=0;
float ErrorS=0;
//Randomize K
void RandomizeK(){
	for(int i=0; i<BackFilter;i++)	{
		Kca[i]=rand()%BackFilter;
 	}
}
// Make the operacion of Result=sum(Vin*Kpar) Remeber that Vin and Kpar are vectors
void ProductoYPrecedir(){
	Result=0;
	for(int i=0; i<BackFilter;i++)	{
		Pro[i]=Vin[i]*Kca[i];
		Result=Pro[i]+Result;
 	}
}
//Initialize the class
void Initialize(){
	ClearData();
	RandomizeK();
	Vin[BackFilter-1]=1; //La ultima posicion es el desface llamada Beta
}
//Print my data of learning
void PrintMySavedData(){
	printf("Saved:\t");
	for(int i=0; i<BackFilter;i++)	{
		printf("%3.3f\t",Vin[i]);
 	}
 	printf("\nK:\t");
	for(int i=0; i<BackFilter;i++)	{
		printf("%3.3f\t",Kca[i]);
 	}
}
//Insert an X value, remember the last position are 1 because
/*
y=mx+b
Where K are a vector that represents m
And Vin are a vector that represents x
But the b can be absorbed with
[x1,x2,x3,...] point [k1,k2,k3] +b => [x1,x2,x3,..,1] point [k1,k2,k3,b]
*/
void AddX(float Add){
	for(int i=0; i<(BackFilter-1);i++)	{
		Vin[i]=Vin[i+1];
 	}
 	Vin[BackFilter-2]=Add;
}
//This traing the model One Time
/*Using 
	h(x)=mx+b
	Where 
	e=1/2*(y-h(x))^2
	de/dk=(y-h(x))*(-Vin) //Remeber that y-h(x) is a number that will be multiplied by a vector
	//(y-h(x))*(-Vin) => This will be a vector
	//k(n)=k(n-1)-(y-h(x))*(-Vin)=> This will be a vector
	k(n)=k(n-1)+(y-h(x))*(Vin) =>Remeber that k is a vector and de/dk also
*/
void TraingOnce(float WanttedData){
	ProductoYPrecedir();
	ErrorN=WanttedData-Result;
	ErrorS=1/2*(ErrorN)*ErrorN;
	for(int i=0; i<BackFilter;i++)	{
		Kca[i]=Kca[i]+ErrorN*Vin[i]*0.1;
 	}
}

}; #To se the code you can download the code, or download the .html pages, the code was developed on DevC++