193 lines
8.2 KiB
C
193 lines
8.2 KiB
C
|
/*
|
||
|
|
LA: linear algebra C++ interface library
|
||
|
|
Kernel ridge regression module Copyright (C) 2024
|
||
|
|
Pavel Florian <florian43@seznam.cz> and Jiri Pittner <jiri.pittner@jh-inst.cas.cz> or <jiri@pittnerovi.com>
|
||
|
|
|
||
|
|
This program is free software: you can redistribute it and/or modify
|
||
|
|
it under the terms of the GNU General Public License as published by
|
||
|
|
the Free Software Foundation, either version 3 of the License, or
|
||
|
|
(at your option) any later version.
|
||
|
|
|
||
|
|
This program is distributed in the hope that it will be useful,
|
||
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
|
|
GNU General Public License for more details.
|
||
|
|
|
||
|
|
You should have received a copy of the GNU General Public License
|
||
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
||
|
|
*/
|
||
|
|
#include <iostream>
|
||
|
|
#include <string>
|
||
|
|
#include <cmath>
|
||
|
|
#include <bits/stdc++.h>
|
||
|
|
#include <fstream>
|
||
|
|
using namespace std;
|
||
|
|
# include "mat.h" // vector support libla implementation
|
||
|
|
# include "vec.h" // compiler parameters: -llapack -lblas
|
||
|
|
|
||
|
|
# include "la.h"
|
||
|
|
# ifndef REG_H
|
||
|
|
# define REG_H
|
||
|
|
|
||
|
|
namespace LA {
|
||
|
|
|
||
|
|
template <typename T>
|
||
|
|
T Gaussian_kernel(const T* x, const unsigned int dimensions, const T sigma);
|
||
|
|
|
||
|
|
template <typename T>
|
||
|
|
T Laplacian_kernel(const T* x, const unsigned int dimensions, const T sigma);
|
||
|
|
|
||
|
|
template <typename T>
|
||
|
|
T Exponential_kernel(const T* x, const unsigned int dimensions, const T sigma);
|
||
|
|
|
||
|
|
template <typename T>
|
||
|
|
T Matern_kernel(const T* x, const unsigned int dimensions, const T sigma, const unsigned int n, const T* Matern_comb_number);
|
||
|
|
|
||
|
|
template <typename T>
|
||
|
|
unsigned int Matern_coefficient(const unsigned int n, T* comb_number);
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
// The kernel regression of points in 1D (curve), 2D (surface), 3D and multi-D space
|
||
|
|
template <typename T>
|
||
|
|
class REG { // Kernel regression (REG) library
|
||
|
|
public:
|
||
|
|
// Constructor of kernel regression model from memory location
|
||
|
|
REG(const T* x, const T* y, const T* params, const unsigned int data_size, const unsigned int x_dimensions, const string kernel);
|
||
|
|
// Constructor of kernel regression model from files
|
||
|
|
REG(const string x_file, const string y_file, const string param_file);
|
||
|
|
// Init function for constructor
|
||
|
|
void Init(const T* x, const T* y, const T* params, const unsigned int data_size, const unsigned int x_dimensions, const string kernel);
|
||
|
|
// General kernel function
|
||
|
|
inline T Kernel(const T* x, const unsigned int dimensions, const unsigned int kernel_id);
|
||
|
|
// Function, that build matrix of distances between kernels
|
||
|
|
unsigned int Build_REG_distances_mat();
|
||
|
|
// Function, that build matrix of coefficients for kernel coordinates
|
||
|
|
unsigned int Build_REG_kernels_mat();
|
||
|
|
// Function, that build sums of matrix coefficients for kernel coordinates
|
||
|
|
unsigned int Build_REG_kernels_mat_sums();
|
||
|
|
// Function, that return root mean square error of model
|
||
|
|
T REG_RMSE(); // sum(delta_y)
|
||
|
|
// The Loss function
|
||
|
|
T Loss_Function(); // sum(delta_y) + lambda((1 - alpha)/2 * sum(weight^2) + alpha * sum(abs(weight)))
|
||
|
|
|
||
|
|
// Function, that learn kernels and set weights
|
||
|
|
unsigned int REG_Fit(const T max_Loss, const unsigned int max_learning_cycles);
|
||
|
|
// Function, that learn kernels and set weights by gradient descent method
|
||
|
|
unsigned int REG_Fit_grad(const T max_Loss, const unsigned int max_learning_cycles, const T learning_rate);
|
||
|
|
// Function, that store fitted parameters of regression model to file
|
||
|
|
unsigned int REG_Save_fitted(const string model_file);
|
||
|
|
// Function, that load fitted parameters of model to file
|
||
|
|
unsigned int REG_Load_fitted(const string model_file);
|
||
|
|
// Function, that uses the kernel regression model to predict y for new data
|
||
|
|
unsigned int REG_Get_predictions(const T* x, T* y_preds, const unsigned int data_size);
|
||
|
|
// Function, that uses the kernel regression model to predict y for new data and investigate the model precision
|
||
|
|
unsigned int REG_Get_Score(const T* x, T* y_preds, const unsigned int data_size, const T* y, T* score);
|
||
|
|
// Setting parameters of kernel regression model
|
||
|
|
unsigned int REG_Set_params(const T* params, const string kernel);
|
||
|
|
// Getting parameters of kernel regression model
|
||
|
|
unsigned int REG_Get_params(T* params, string* kernel);
|
||
|
|
// Getting weight coefficients of kernels in regression model
|
||
|
|
unsigned int REG_Get_weight_coeff(T* weight);
|
||
|
|
// destructor of kernel regression model
|
||
|
|
~REG();
|
||
|
|
|
||
|
|
private:
|
||
|
|
// string with type of kernel
|
||
|
|
string REG_kernel_type;
|
||
|
|
// number of type of kernel
|
||
|
|
unsigned int REG_kernel_id;
|
||
|
|
// Count of x items (dimensions) for kernels
|
||
|
|
unsigned int REG_count_dimensions;
|
||
|
|
// number of input x values/kernels
|
||
|
|
unsigned int REG_count_kernels;
|
||
|
|
// lambda parameter of regression
|
||
|
|
T REG_lambda;
|
||
|
|
// alpha parameter of regression
|
||
|
|
T REG_alpha;
|
||
|
|
// sigma parameter of kernels
|
||
|
|
T REG_sigma;
|
||
|
|
// n for Matern kernels
|
||
|
|
unsigned int REG_n;
|
||
|
|
// indicator of initialization or initialization errors
|
||
|
|
bool REG_base_init = false;
|
||
|
|
// indicator of initialization of allocation of memory
|
||
|
|
bool REG_aloc_init = false;
|
||
|
|
// indicator of initialization of REG_REG_distances_mat and REG_REG_kernels_mat
|
||
|
|
bool REG_mat_init = false;
|
||
|
|
|
||
|
|
// Matern kernel coefficient computed from n and k
|
||
|
|
T* REG_Matern_comb_number = nullptr;
|
||
|
|
// pointer to parameters of kernel regression, various size, first is lambda, second alpha, third sigma parameter of kernels
|
||
|
|
// and fourth the n parameter, if is used Matern kernel
|
||
|
|
T* REG_params = nullptr;
|
||
|
|
// variable which contains number of learning cycles of model
|
||
|
|
unsigned int learning_cycles = 0;
|
||
|
|
|
||
|
|
// x values of kernel regression input data
|
||
|
|
T* REG_coordinates = nullptr;
|
||
|
|
// y values of kernel regression input data
|
||
|
|
T* REG_Y = nullptr;
|
||
|
|
|
||
|
|
// matrix of distances between kernels
|
||
|
|
T* REG_distances = nullptr;
|
||
|
|
// matrix of kernel coefficients for kernel coordinates
|
||
|
|
T* REG_kernels = nullptr;
|
||
|
|
// vector of row sums of kernel coefficients for kernel coordinates
|
||
|
|
T* REG_kernels_sums = nullptr;
|
||
|
|
// vector of weight coefficients for each kernel
|
||
|
|
T* REG_weights = nullptr;
|
||
|
|
|
||
|
|
// vector of distances between kernels
|
||
|
|
T* REG_coordinates_diff = nullptr;
|
||
|
|
// vector of kernel coefficients for kernel coordinates
|
||
|
|
T* REG_kernels_diff = nullptr;
|
||
|
|
// vector of y predictions for each kernel
|
||
|
|
T* REG_y_preds = nullptr;
|
||
|
|
// vector of differences between y values and predictions for each kernel
|
||
|
|
T* REG_y_delta = nullptr;
|
||
|
|
// vector of temporary results for processing y values and predictions for each kernel
|
||
|
|
T* REG_y_temp = nullptr;
|
||
|
|
|
||
|
|
// pointer to x buffer for model loaded from file
|
||
|
|
T* REG_x_file = nullptr;
|
||
|
|
// pointer to y buffer for model loaded from file
|
||
|
|
T* REG_y_file = nullptr;
|
||
|
|
// pointer to parameters buffer for model loaded from file
|
||
|
|
T* REG_param_file = nullptr;
|
||
|
|
// pointer to fitted data for model loaded from file
|
||
|
|
T* REG_fit_file = nullptr;
|
||
|
|
|
||
|
|
// NRVec for Matern kernel coefficient computed from n and k
|
||
|
|
NRVec<T> REG_Matern_comb_number_vec;
|
||
|
|
|
||
|
|
// NRVec for matrix of distances between kernels
|
||
|
|
NRVec<T> REG_coordinates_vec;
|
||
|
|
// NRVec for vector of y imput values of kernels
|
||
|
|
NRVec<T> REG_y_vec;
|
||
|
|
// NRMat for matrix of distances between kernels
|
||
|
|
NRMat<T> REG_distances_mat;
|
||
|
|
// NRMat for matrix of kernel coefficients for kernel coordinates
|
||
|
|
NRMat<T> REG_kernels_mat;
|
||
|
|
// NRVec for vector of row sums of kernel coefficients for kernel coordinates
|
||
|
|
NRVec<T> REG_kernels_sums_vec;
|
||
|
|
// NRVec for vector of weight coefficients of kernels
|
||
|
|
NRVec<T> REG_weights_vec;
|
||
|
|
|
||
|
|
// NRVec for matrix of coordinates differences between kernels
|
||
|
|
NRVec<T> REG_coordinates_diff_vec;
|
||
|
|
// NRVec for matrix of kernel coefficients for kernel coordinates
|
||
|
|
NRVec<T> REG_kernels_diff_vec;
|
||
|
|
|
||
|
|
// NRVec for vector of y predictions of kernels
|
||
|
|
NRVec<T> REG_y_preds_vec;
|
||
|
|
// NRVec for vector of differences between estimated real and real y values for training kernels
|
||
|
|
NRVec<T> REG_y_delta_vec;
|
||
|
|
// NRVec for temporary results for processing y values and predictions for each kernel
|
||
|
|
NRVec<T> REG_y_temp_vec;
|
||
|
|
};
|
||
|
|
// Include the template classes definitions
|
||
|
|
|
||
|
|
} // end of namespace
|
||
|
|
# endif /* REG_H */
|