mirror of
https://github.com/augustin64/projet-tipe
synced 2025-01-24 15:36:25 +01:00
503 lines
16 KiB
C
503 lines
16 KiB
C
|
/*
|
||
|
Version du module Matrice avec des matrices 2d
|
||
|
Une version 3d doit s'inspirer de celle-ci
|
||
|
*/
|
||
|
|
||
|
|
||
|
|
||
|
|
||
|
|
||
|
#include <stdint.h>
|
||
|
#include <stdlib.h>
|
||
|
#include <stdio.h>
|
||
|
#include <float.h>
|
||
|
#include <math.h>
|
||
|
|
||
|
|
||
|
typedef struct Matrix {
|
||
|
int rows; // Nombre de lignes de la matrice
|
||
|
int columns; // Nombre de colonnes de la matrice
|
||
|
float** value; // Tableau 2d comportant les valeurs de matrice
|
||
|
|
||
|
} Matrix;
|
||
|
|
||
|
|
||
|
|
||
|
|
||
|
float exp_float(float a);
|
||
|
float max_float(float a, float b);
|
||
|
float min_float(float a, float b);
|
||
|
Matrix* create_matrix(int nb_rows, int nb_columns);
|
||
|
void uniformity_matrix(Matrix* m, float v);
|
||
|
void print_matrix(Matrix* m);
|
||
|
float number_from_matrix(Matrix* m);
|
||
|
float max_in_matrix(Matrix* m);
|
||
|
void free_matrix(Matrix* m);
|
||
|
void product_of_a_scalar_matrix(Matrix* m, float scalar);
|
||
|
void sum_of_a_scalar_matrix(Matrix* m, float scalar);
|
||
|
Matrix* new_copy_matrix(Matrix* m);
|
||
|
Matrix* apply_function_new_matrix(Matrix* m, float (*f)(float));
|
||
|
void apply_function_matrix(Matrix* m, float (*f)(float));
|
||
|
void transpose_matrix(Matrix* m);
|
||
|
void add_matrix(Matrix* m1, Matrix* m2);
|
||
|
Matrix* product_matrix(Matrix* m1, Matrix* m2);
|
||
|
void max_pooling_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out);
|
||
|
void min_pooling_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out);
|
||
|
void average_pooling_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out);
|
||
|
void valid_cross_correlation_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out);
|
||
|
void full_cross_correlation_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out);
|
||
|
void softmax_matrix(Matrix* m);
|
||
|
float quadratic_cost_matrix(Matrix* m, int i_number, int j_number);
|
||
|
void rotation_180_matrix(Matrix* m);
|
||
|
|
||
|
|
||
|
|
||
|
float exp_float(float a) {
|
||
|
/* Renvoie l'exponentiel d'un flotant '*/
|
||
|
return (float)exp(a);
|
||
|
}
|
||
|
|
||
|
|
||
|
float max_float(float a, float b) {
|
||
|
/* Renvoie le max entre les deux flotants */
|
||
|
return a>b?a:b;
|
||
|
}
|
||
|
|
||
|
|
||
|
float min_float(float a, float b) {
|
||
|
/* Renvoie le min entre les deux flotants */
|
||
|
return a<b?a:b;
|
||
|
}
|
||
|
|
||
|
|
||
|
Matrix* create_matrix(int nb_rows, int nb_columns) {
|
||
|
/* Créé une matrice en lui allouant de la mémoire */
|
||
|
Matrix* m = malloc(sizeof(Matrix));
|
||
|
m->rows = nb_rows;
|
||
|
m->columns = nb_columns;
|
||
|
m->value = malloc(sizeof(float*)*m->rows);
|
||
|
for (int i=0; i < m->rows; i++)
|
||
|
m->value[i] = malloc(sizeof(float)*m->columns);
|
||
|
return m;
|
||
|
}
|
||
|
|
||
|
|
||
|
void uniformity_matrix(Matrix* m, float v) {
|
||
|
/* Insère la même valeur partout dans la matrice */
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
m->value[i][j] = v;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void print_matrix(Matrix* m) {
|
||
|
/* Affiche la matrice */
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
if (j!=0)
|
||
|
printf(",");
|
||
|
printf("%f ", m->value[i][j]);
|
||
|
}
|
||
|
printf("\n");
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
float number_from_matrix(Matrix* m) {
|
||
|
/* Renvoie la somme des éléments de la matrice */
|
||
|
float tmp=0;
|
||
|
for (int i=0; i < m->rows ; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
tmp += m->value[i][j];
|
||
|
}
|
||
|
}
|
||
|
return tmp;
|
||
|
}
|
||
|
|
||
|
|
||
|
float max_in_matrix(Matrix* m) {
|
||
|
/* Renvoie l'élément maximal de la matrice */
|
||
|
float max_tmp = FLT_MIN;
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
max_tmp = max_float(max_tmp, m->value[i][j]);
|
||
|
}
|
||
|
}
|
||
|
return max_tmp;
|
||
|
}
|
||
|
|
||
|
|
||
|
void free_matrix(Matrix* m) {
|
||
|
/* Libère l'espace mémoire alloué à la matrice */
|
||
|
for (int i=0; i < m->rows; i++)
|
||
|
free(m->value[i]);
|
||
|
free(m->value);
|
||
|
}
|
||
|
|
||
|
|
||
|
void product_of_a_scalar_matrix(Matrix* m, float scalar) {
|
||
|
/* Multiplie la matrice par un scalaire */
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
m->value[i][j] *= scalar;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void sum_of_a_scalar_matrix(Matrix* m, float scalar) {
|
||
|
/* Ajoute un scalaire à la matrice */
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
m->value[i][j] += scalar;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
Matrix* new_copy_matrix(Matrix* m) {
|
||
|
/* Renvoie une copie de la matrice */
|
||
|
Matrix* new_m = create_matrix(m->rows, m->columns);
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
new_m->value[i][j] = m->value[i][j];
|
||
|
}
|
||
|
}
|
||
|
return new_m;
|
||
|
}
|
||
|
|
||
|
|
||
|
void copy_matrix(Matrix* m1, Matrix* m2) {
|
||
|
/* Copie le contenu de la matrice m1 dans la matrice m2 */
|
||
|
if (m1->rows != m2->rows || m1->columns != m2->columns) {
|
||
|
printf("Erreur, copie dans de deux matrices dont les dimensions diffèrent");
|
||
|
return;
|
||
|
}
|
||
|
for (int i=0; i < m1->rows; i++) {
|
||
|
for (int j=0; j < m2->columns; j++) {
|
||
|
m2->value[i][j] = m1->value[i][j];
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
Matrix* apply_function_new_matrix(Matrix* m, float (*f)(float)) {
|
||
|
/* Renvoie une matrice avec une fonction appliquée
|
||
|
à tous les éléments de l'ancienne matrice */
|
||
|
Matrix* new_m = create_matrix(m->rows, m->columns);
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m ->columns; j++) {
|
||
|
new_m->value[i][j] = (*f)(m->value[i][j]);
|
||
|
}
|
||
|
}
|
||
|
return new_m;
|
||
|
}
|
||
|
|
||
|
|
||
|
void apply_function_matrix(Matrix* m, float (*f)(float)) {
|
||
|
/* Applique une fonction à tous les éléments de la matrice */
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m ->columns; j++) {
|
||
|
m->value[i][j] = (*f)(m->value[i][j]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void transpose_matrix(Matrix* m) {
|
||
|
/* Transpose la matrice si c'est possible */
|
||
|
if (m->rows != m->columns) {
|
||
|
printf("Erreur, matrice non compatible avec la transposition");
|
||
|
return;
|
||
|
}
|
||
|
float cpt;
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=i+1; j < m->columns; j++) {
|
||
|
cpt = m->value[i][j];
|
||
|
m->value[i][j] = m->value[j][i];
|
||
|
m->value[j][i] = cpt;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void add_matrix(Matrix* m1, Matrix* m2) {
|
||
|
/* Ajoute la matrice m1 à la matrice m2 */
|
||
|
if (m1->rows != m2->rows || m1->columns != m2->columns) {
|
||
|
printf("Erreur, matrices non compatibles avec la somme");
|
||
|
return;
|
||
|
}
|
||
|
for (int i=0; i < m2->rows; i++) {
|
||
|
for (int j=0; j < m2->columns; j++) {
|
||
|
m2->value[i][j] += m1->value[i][j];
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
Matrix* product_matrix(Matrix* m1, Matrix* m2) {
|
||
|
/* Renvoie une nouvelle matrice produit (classique)
|
||
|
des deux matrices si les dimensions sont correctes*/
|
||
|
if (m1->columns != m2->rows) {
|
||
|
printf("Erreur, matrices non compatibles avec le produit");
|
||
|
return NULL;
|
||
|
}
|
||
|
float cpt;
|
||
|
Matrix* m = create_matrix(m1->rows, m2->columns);
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
cpt=0;
|
||
|
for (int k=0; k < m2->rows; k++) {
|
||
|
cpt += m1->value[i][j]* m2->value[k][j];
|
||
|
}
|
||
|
m->value[i][j] = cpt;
|
||
|
}
|
||
|
}
|
||
|
return m;
|
||
|
}
|
||
|
|
||
|
|
||
|
void max_pooling_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out) {
|
||
|
/* Insère le résultat de max pooling avec un décalage
|
||
|
de (stride) pixels dans la matrice m_out */
|
||
|
if (m_in->columns < kernel->columns || m_in->rows < kernel->rows) {
|
||
|
printf("Erreur, kernel plus grand que la matrice dans max pooling");
|
||
|
return;
|
||
|
}
|
||
|
if (((m_in->columns - kernel->columns)/stride)+1 != m_out->columns || ((m_in->rows - kernel->rows)/stride)+1 != m_out->rows) {
|
||
|
printf("Erreur, matrice et kernel non compatibles avec le décalage ou la matrice sortante dans max pooling");
|
||
|
return;
|
||
|
}
|
||
|
int i, j, a ,b;
|
||
|
float tmp;
|
||
|
for (i=0; i < m_out->rows; i++) {
|
||
|
for (j=0; j < m_out->columns; j++) {
|
||
|
tmp = FLT_MIN;
|
||
|
for (a=0; a < kernel->rows; a++) {
|
||
|
for (b=0; b < kernel->columns; b++) {
|
||
|
tmp = max_float(tmp, m_in->value[i*stride +a][j*stride +b]);
|
||
|
}
|
||
|
}
|
||
|
m_out->value[i][j] = tmp;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void min_pooling_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out) {
|
||
|
/* Insère le résultat de min pooling avec un décalage
|
||
|
de (stride) pixels dans la matrice m_out */
|
||
|
if (m_in->columns < kernel->columns || m_in->rows < kernel->rows) {
|
||
|
printf("Erreur, kernel plus grand que la matrice dans min pooling");
|
||
|
return;
|
||
|
}
|
||
|
if (((m_in->columns - kernel->columns)/stride)+1 != m_out->columns || ((m_in->rows - kernel->rows)/stride)+1 != m_out->rows) {
|
||
|
printf("Erreur, matrice et kernel non compatibles avec le décalage ou la matrice sortante dans min pooling");
|
||
|
return;
|
||
|
}
|
||
|
int i, j, a ,b;
|
||
|
float tmp;
|
||
|
for (i=0; i < m_out->rows; i++) {
|
||
|
for (j=0; j < m_out->columns; j++) {
|
||
|
tmp = FLT_MAX;
|
||
|
for (a=0; a < kernel->rows; a++) {
|
||
|
for (b=0; b < kernel->columns; b++) {
|
||
|
tmp = min_float(tmp, m_in->value[i*stride +a][j*stride +b]);
|
||
|
}
|
||
|
}
|
||
|
m_out->value[i][j] = tmp;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void average_pooling_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out) {
|
||
|
/* Insère le résultat de max pooling avec un décalage
|
||
|
de (stride) pixels dans la matrice m_out */
|
||
|
if (m_in->columns < kernel->columns || m_in->rows < kernel->rows) {
|
||
|
printf("Erreur, kernel plus grand que la matrice dans average pooling");
|
||
|
return;
|
||
|
}
|
||
|
if (((m_in->columns - kernel->columns)/stride)+1 != m_out->columns || ((m_in->rows - kernel->rows)/stride)+1 != m_out->rows) {
|
||
|
printf("Erreur, matrice et kernel non compatibles avec le décalage ou la matrice sortante dans average pooling");
|
||
|
return;
|
||
|
}
|
||
|
int i, j, a, b, nb= kernel->rows*kernel->columns;
|
||
|
for (i=0; i < m_out->rows; i++) {
|
||
|
for (j=0; j < m_out->columns; j++) {
|
||
|
m_out->value[i][j] = 0;
|
||
|
for (a=0; a < kernel->rows; a++) {
|
||
|
for (b=0; b < kernel->columns; b++) {
|
||
|
m_out->value[i][j] += m_in->value[i*stride +a][j*stride +b];
|
||
|
}
|
||
|
}
|
||
|
m_out->value[i][j] = m_out->value[i][j]/nb;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void valid_cross_correlation_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out) {
|
||
|
/* Ajoute, la cross-correlation valide de m_in et
|
||
|
kernel avec un décalage de stride, dans m_out */
|
||
|
if (m_in->columns < kernel->columns || m_in->rows < kernel->rows) {
|
||
|
printf("Erreur, kernel plus grand que la matrice dans valid cross-correlation");
|
||
|
return;
|
||
|
}
|
||
|
if (((m_in->columns - kernel->columns)/stride)+1 != m_out->columns || ((m_in->rows - kernel->rows)/stride)+1 != m_out->rows) {
|
||
|
printf("Erreur, matrice et kernel non compatibles avec le décalage ou la matrice sortante dans valid cross-correlation");
|
||
|
return;
|
||
|
}
|
||
|
int i, j, a, b;
|
||
|
for (i=0; i < m_out->rows; i++) {
|
||
|
for (j=0; j < m_out->columns; j++) {
|
||
|
for (a=0; a < kernel->rows; a++) {
|
||
|
for (b=0; b < kernel->columns; b++) {
|
||
|
m_out->value[i][j] += m_in->value[i*stride +a][j*stride +b]*kernel->value[a][b];
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void full_cross_correlation_matrix(Matrix* m_in, Matrix* kernel, int stride, Matrix* m_out) {
|
||
|
/* Ajoute, la cross-correlation entière de m_in et
|
||
|
kernel avec un décalage de stride, dans m_out */
|
||
|
int rows_k = kernel->rows-1;
|
||
|
int columns_k = kernel->columns-1;
|
||
|
if (m_in->columns < kernel->columns || m_in->rows < kernel->rows) {
|
||
|
printf("Erreur, kernel plus grand que la matrice dans full cross-correlation");
|
||
|
return;
|
||
|
}
|
||
|
if ((m_in->columns + 2*columns_k)/stride != m_out->columns || (m_in->rows + 2*rows_k)/stride != m_out->rows) {
|
||
|
printf("Erreur, matrice et kernel non compatibles avec le décalage ou la matrice sortante dans full cross-correlation");
|
||
|
return;
|
||
|
}
|
||
|
int i, j, a, b, new_i, new_j;
|
||
|
for (i=-rows_k; i < (m_out->rows + kernel->rows -1); i++) {
|
||
|
for (j=-columns_k; j < (m_out->columns + kernel->columns -1); j++) {
|
||
|
m_out->value[i+rows_k][j+columns_k] = 0;
|
||
|
for (a=0; a < kernel->rows; a++) {
|
||
|
for (b=0; b < kernel->columns; b++) {
|
||
|
new_i = i*stride +a;
|
||
|
new_j = j*stride +b;
|
||
|
if (new_i >= 0 || new_i < m_in->rows || new_j >= 0 || new_j < m_in->columns)
|
||
|
m_out->value[i+rows_k][j+columns_k] += m_in->value[i*stride +a][j*stride +b]*kernel->value[a][b];
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void softmax_matrix(Matrix* m) {
|
||
|
/* Applique la fonction softmax sur la matrice en changeant ses valeurs */
|
||
|
float max = max_in_matrix(m);
|
||
|
sum_of_a_scalar_matrix(m, (-1)*max);
|
||
|
apply_function_matrix(m, exp_float);
|
||
|
float sum = number_from_matrix(m);
|
||
|
sum = 1/sum;
|
||
|
product_of_a_scalar_matrix(m, sum);
|
||
|
}
|
||
|
|
||
|
|
||
|
float quadratic_cost_matrix(Matrix* m, int i_number, int j_number) {
|
||
|
/* Renvoie l'erreur de la matrice où les valeurs sont des probabailités */
|
||
|
float loss = 0;
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=0; j < m->columns; j++) {
|
||
|
if (i==i_number && j==j_number)
|
||
|
loss += (1-m->value[i][j])*(1-m->value[i][j]);
|
||
|
else
|
||
|
loss += m->value[i][j]*m->value[i][j];
|
||
|
}
|
||
|
}
|
||
|
return loss;
|
||
|
}
|
||
|
|
||
|
|
||
|
void rotation_180_matrix(Matrix* m) {
|
||
|
/* Modifie la matrice en pivotant ses valeurs de 180° */
|
||
|
if (m->rows != m-> columns) {
|
||
|
printf("Erreur, une matrice non carrée ne peut pas être retourner");
|
||
|
return;
|
||
|
}
|
||
|
float tmp;
|
||
|
int half_rows = m->rows/2;
|
||
|
int max_r = m->rows-1;
|
||
|
int max_c = m->columns-1;
|
||
|
for (int i=0; i < m->rows; i++) {
|
||
|
for (int j=i; j < m->columns; j++) {
|
||
|
if (i!=j || i>=half_rows) {
|
||
|
tmp = m->value[i][j];
|
||
|
m->value[i][j] = m->value[max_r-i][max_c-j];
|
||
|
m->value[max_r-i][max_c-j] = tmp;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void valid__cross_correlation_step_forward(Matrix** layer_input, Matrix*** layer_kernel, Matrix** layer_bias, Matrix** layer_output, int len_layer, int depth_kernel, int stride) {
|
||
|
/* Effectue une étape de la forward-propagation
|
||
|
à l'aide d'une cross-correlation valide */
|
||
|
for (int i=0; i < depth_kernel; i++) {
|
||
|
copy_matrix(layer_bias[i], layer_output[i]);
|
||
|
|
||
|
for (int j=0; j < len_layer; j++) {
|
||
|
valid_cross_correlation_matrix(layer_input[j], layer_kernel[i][j], stride, layer_output[j]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void max_pooling_step_forward(Matrix** layer_input, Matrix*** layer_kernel, Matrix** layer_bias, Matrix** layer_output, int len_layer, int depth_kernel, int stride) {
|
||
|
/* Effectue une étape de la forward-propagation
|
||
|
à l'aide d'un max_pooling */
|
||
|
for (int i=0; i < depth_kernel; i++) {
|
||
|
copy_matrix(layer_bias[i], layer_output[i]);
|
||
|
|
||
|
for (int j=0; j < len_layer; j++) {
|
||
|
max_pooling_matrix(layer_input[j], layer_kernel[i][j], stride, layer_output[j]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void average_pooling_step_forward(Matrix** layer_input, Matrix*** layer_kernel, Matrix** layer_bias, Matrix** layer_output, int len_layer, int depth_kernel, int stride) {
|
||
|
/* Effectue une étape de la forward-propagation
|
||
|
à l'aide d'un average_pooling */
|
||
|
for (int i=0; i < depth_kernel; i++) {
|
||
|
copy_matrix(layer_bias[i], layer_output[i]);
|
||
|
|
||
|
for (int j=0; j < len_layer; j++) {
|
||
|
average_pooling_matrix(layer_input[j], layer_kernel[i][j], stride, layer_output[j]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
void reshape_step_forward(Matrix** layer_input, Matrix*** layer_kernel, Matrix** layer_bias, Matrix** layer_output, int len_layer, int depth_kernel, int stride) {
|
||
|
/* Effectue une étape de la forward-propagation
|
||
|
en redimensionnant la matrice */
|
||
|
for (int i=0; i < depth_kernel; i++) {
|
||
|
copy_matrix(layer_bias[i], layer_output[i]);
|
||
|
|
||
|
for (int j=0; j < len_layer; j++) {
|
||
|
average_pooling_matrix(layer_input[j], layer_kernel[i][j], stride, layer_output[j]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
|
||
|
int main() {
|
||
|
Matrix* m = create_matrix(2, 2);
|
||
|
uniformity_matrix(m, 1);
|
||
|
print_matrix(m);
|
||
|
free_matrix(m);
|
||
|
return 0;
|
||
|
}
|