MotorControlModuleSDFM_TMS320F28388D/Projects/EFC_Communication/UMLibrary/algorithm/LinearRegression.cpp

/*
 * LinearRegression.cpp
 *
 *  Created on: 23 нояб. 2016 г.
 *      Author: titov
 */

#include "LinearRegression.hh"

#include <cmath>

using algorithm::LinearRegression;

LinearRegression::LinearRegression(unsigned int init_rowData, unsigned int init_colData,  std::pmr::memory_resource * memory ) :
        memory(memory) {

	colData = init_colData;
	rowData = init_rowData;

	alfa = 0;
	startCalculation = false;
	numCalculation = 0;

	data = std::pmr::polymorphic_allocator<float *>(memory).allocate(rowData);//new float * [rowData];
	for (int i = 0; i < rowData; i++)
		data[i] = std::pmr::polymorphic_allocator<float>(memory).allocate(colData);//new float [colData];

	Theta = std::pmr::polymorphic_allocator<float>(memory).allocate(colData);//new float [colData];
	TempTheta = std::pmr::polymorphic_allocator<float>(memory).allocate(colData);//new float [colData];
	result = std::pmr::polymorphic_allocator<float>(memory).allocate(rowData);//new float [rowData];

	for( int j = 0; j < colData; j++ ) {

	    Theta[j] = 0.0f;
	    TempTheta[j] = 0.0f;

	}


	rowIndex = 0;
	colIndex = 0;

	for( int i = 0; i < rowData; i++ ) 	{
		for( int j = 0; j < colData; j++ )
			data[i][j] = NAN;

		result[i] = NAN;
	}
}

LinearRegression::~LinearRegression()
{
	for( int i = 0; i < rowData; i++ )
	    std::pmr::polymorphic_allocator<float>(memory).deallocate( data[i], colData );//delete [] data[i];

	std::pmr::polymorphic_allocator<float *>(memory).deallocate( data, rowData );// delete [] data;

	std::pmr::polymorphic_allocator<float>(memory).deallocate( Theta, colData );//delete [] Theta;
	std::pmr::polymorphic_allocator<float>(memory).deallocate( TempTheta, colData );//delete [] TempTheta;
	std::pmr::polymorphic_allocator<float>(memory).deallocate( result, rowData );//delete [] result;
}

//float algorithm::LinearRegression::errorCalc(float hp, float res) const {
//
//    return hp - res;
//
//}

float LinearRegression::getTheta(unsigned int idTheta) const
{
	if(idTheta < colData)
		return Theta[idTheta];
	else
		return NAN;
}

bool LinearRegression::addData(float * new_data, float new_result)
{
	if(rowIndex < rowData)
	{
		for(int i = 0; i < colData; i++)
		{
			data[rowIndex][i] = new_data[i];
		}

		result[rowIndex] = new_result;

		rowIndex++;
	}
	else
	{
		return false;
	}

	return true;
}

bool LinearRegression::init(float initAlfa, float * initTheta, unsigned int numCalc)
{
	alfa = initAlfa;
	startCalculation = false;
	numCalculation = numCalc;

	bool created;

	if(Theta && TempTheta && result && data) {
		created = true;
		for(int i = 0; i < rowData; i++) {
			created = created && data[i];
		}
	} else {
		created = false;
	}

	for(int i = 0; i < colData && created; i++)
		Theta[i] = initTheta[i];

	return created;
}

bool LinearRegression::update()
{
	float h_p = 0.0f;
	float update_Theta = 0.0f;
	float delta_Error = 0.0f;

	for(int k = 0; k < colData; k++)					//Расчитываем обновление для каждого значение Theta
	{
		update_Theta = 0.0f;
		for(int j = 0; j < rowData; j++)				//Считаем суммарную ошибку.
		{
			h_p = 0.0f;
			for(int i = 0; i < colData; i++)			//Вычислчем гипотезу.
				h_p += Theta[i] * data[j][i];

			delta_Error = this->errorCalc(h_p, result[j]);	//Вычисляем ошибку.
			update_Theta += delta_Error * data[j][k];	//Вычисляем обновляющее значение.
		}

		TempTheta[k] = Theta[k] - alfa * (update_Theta / rowData);
	}

	for(int n = 0; n < colData; n++) 					//Обновляем значения Theta.
	{
		Theta[n] = TempTheta[n];
	}

	return true;
}

bool LinearRegression::calculation()
{
	if(startCalculation && numCalculation)
	{
		update();
		numCalculation--;
		return true;
	} else {
		return false;
	}
}

float algorithm::LinearRegressionCirc::errorCalc(float hp, float res) const {

    float error = hp - res;

    if(error > math::constants::pi) {           //Выбираем наименьшее значение длинны, так как находимся на окружности.
        error = error - math::constants::pi2;
    } else if (error <= -math::constants::pi) {
        error = error + math::constants::pi2;
    }

    return error;

}

float algorithm::LinearRegression::getError() const {

    float h_p = 0.0f;
    float update_Theta = 0.0f;
    float delta_Error = 0.0f;

    for(int k = 0; k < colData; k++)                    //Расчитываем обновление для каждого значение Theta
    {
        update_Theta = 0.0f;
        for(int j = 0; j < rowData; j++)                //Считаем суммарную ошибку.
        {
            h_p = 0.0f;
            for(int i = 0; i < colData; i++)            //Вычислчем гипотезу.
                h_p += Theta[i] * data[j][i];

            delta_Error += this->errorCalc(h_p, result[j]);  //Вычисляем ошибку.
        }

    }

    return delta_Error / rowData;
}