379 lines
No EOL
12 KiB
C
379 lines
No EOL
12 KiB
C
#include <stdlib.h>
|
|
#include "neuronal_network.h"
|
|
#include <stdio.h>
|
|
#include <time.h>
|
|
#include <math.h>
|
|
|
|
double sigmoid(double input);
|
|
double sigmoid_derivative(double x);
|
|
|
|
Matrix* softmax(Matrix* matrix);
|
|
double square(double input);
|
|
|
|
double loss_function(Matrix* output_matrix, int image_label);
|
|
|
|
void backPropagation(double learning_rate, Matrix* weights, Matrix* biases, Matrix* current_layer_activation, Matrix* previous_layer_activation, Matrix* sigma_old);
|
|
|
|
Neural_Network* new_network(int input_size, int hidden_size, int output_size, double learning_rate){
|
|
Neural_Network *network = malloc(sizeof(Neural_Network));
|
|
// initialize networks variables
|
|
network->hidden_size = hidden_size;
|
|
network->input_size = input_size;
|
|
network->output_size = output_size;
|
|
network->learning_rate = learning_rate;
|
|
|
|
network->weights_1 = matrix_create(hidden_size, input_size);
|
|
network->weights_2 = matrix_create(hidden_size, hidden_size);
|
|
network->weights_3 = matrix_create(hidden_size, hidden_size);
|
|
network->weights_output = matrix_create(output_size, hidden_size);
|
|
network->bias_1 = matrix_create(hidden_size, 1);
|
|
network->bias_2 = matrix_create(hidden_size, 1);
|
|
network->bias_3 = matrix_create(hidden_size, 1);
|
|
network->bias_output = matrix_create(output_size, 1);
|
|
|
|
return network;
|
|
}
|
|
|
|
void randomize_network(Neural_Network* network, int scope){
|
|
matrix_randomize(network->weights_1, scope);
|
|
matrix_randomize(network->weights_2, scope);
|
|
matrix_randomize(network->weights_3, scope);
|
|
matrix_randomize(network->weights_output, scope);
|
|
matrix_randomize(network->bias_1, scope);
|
|
matrix_randomize(network->bias_2, scope);
|
|
matrix_randomize(network->bias_3, scope);
|
|
matrix_randomize(network->bias_output, scope);
|
|
}
|
|
|
|
//void print_network(Neural_Network* network){};
|
|
|
|
void free_network(Neural_Network* network){
|
|
matrix_free(network->weights_1);
|
|
matrix_free(network->weights_2);
|
|
matrix_free(network->weights_3);
|
|
matrix_free(network->weights_output);
|
|
matrix_free(network->bias_1);
|
|
matrix_free(network->bias_2);
|
|
matrix_free(network->bias_3);
|
|
matrix_free(network->bias_output);
|
|
free(network);
|
|
}
|
|
|
|
void save_network(Neural_Network* network) {
|
|
|
|
char* file_name = "../networks/newest_network.txt";
|
|
|
|
// create file
|
|
FILE* save_file = fopen(file_name, "w");
|
|
|
|
// check if file is successfully opened
|
|
if(save_file == NULL) {
|
|
printf("ERROR: Something went wrong in file creation! (save_network)");
|
|
exit(1);
|
|
}
|
|
|
|
// save network size to first line of the file
|
|
fprintf(save_file, "%d\n", network->input_size);
|
|
fprintf(save_file, "%d\n", network->hidden_size);
|
|
fprintf(save_file, "%d\n", network->output_size);
|
|
|
|
// close the file
|
|
fclose(save_file);
|
|
|
|
// save first layer
|
|
matrix_save(network->bias_1, file_name);
|
|
matrix_save(network->weights_1, file_name);
|
|
|
|
// save second layer
|
|
matrix_save(network->bias_2, file_name);
|
|
matrix_save(network->weights_2, file_name);
|
|
|
|
// save third layer
|
|
matrix_save(network->bias_3, file_name);
|
|
matrix_save(network->weights_3, file_name);
|
|
|
|
// save output weights
|
|
matrix_save(network->bias_output, file_name);
|
|
matrix_save(network->weights_output, file_name);
|
|
|
|
printf("Network Saved!");
|
|
}
|
|
|
|
Neural_Network* load_network(char* file) {
|
|
|
|
// create file pointer and open file
|
|
FILE* save_file = fopen(file, "r");
|
|
|
|
// check if file could be opened
|
|
if(save_file == NULL) {
|
|
printf("ERROR: File could not be opened/found! (load_network)");
|
|
exit(1);
|
|
}
|
|
|
|
// read & store the information on the size of the network from the save file
|
|
char buffer[MAX_BYTES];
|
|
fgets(buffer, MAX_BYTES, save_file);
|
|
int input_size = (int) strtol(buffer, NULL, 10);
|
|
fgets(buffer, MAX_BYTES, save_file);
|
|
int hidden_size = (int) strtol(buffer, NULL, 10);
|
|
fgets(buffer, MAX_BYTES, save_file);
|
|
int output_size = (int) strtol(buffer, NULL, 10);
|
|
|
|
// create a new network to fill with the saved data
|
|
Neural_Network* saved_network = new_network(input_size, hidden_size, output_size, 0);
|
|
|
|
// load matrices from file into struct
|
|
saved_network->bias_1 = load_next_matrix(save_file);
|
|
saved_network->weights_1 = load_next_matrix(save_file);
|
|
saved_network->bias_2 = load_next_matrix(save_file);
|
|
saved_network->weights_2 = load_next_matrix(save_file);
|
|
saved_network->bias_3 = load_next_matrix(save_file);
|
|
saved_network->weights_3 = load_next_matrix(save_file);
|
|
saved_network->bias_output = load_next_matrix(save_file);
|
|
saved_network->weights_output = load_next_matrix(save_file);
|
|
|
|
// return saved network
|
|
fclose(save_file);
|
|
return saved_network;
|
|
}
|
|
|
|
double measure_network_accuracy(Neural_Network* network, Image** images, int amount) {
|
|
int num_correct = 0;
|
|
for (int i = 0; i < amount; i++) {
|
|
Matrix* prediction = predict_image(network, images[i]);
|
|
if (matrix_argmax(prediction) == images[i]->label) {
|
|
num_correct++;
|
|
}
|
|
matrix_free(prediction);
|
|
}
|
|
return 1.0 * num_correct / amount;
|
|
}
|
|
|
|
Matrix* predict_image(Neural_Network* network, Image* image){
|
|
Matrix* image_data = matrix_flatten(image->pixel_values, 0);
|
|
Matrix* res = predict(network, image_data);
|
|
matrix_free(image_data);
|
|
return res;
|
|
}
|
|
|
|
Matrix* predict(Neural_Network* network, Matrix* image_data) {
|
|
Matrix* h1_dot = dot(network->weights_1, image_data);
|
|
Matrix* h1_add = add(h1_dot, network->bias_1);
|
|
Matrix* h1_outputs = apply(sigmoid, h1_add);
|
|
|
|
Matrix* h2_dot = dot(network->weights_2, h1_outputs);
|
|
Matrix* h2_add = add(h2_dot, network->bias_2);
|
|
Matrix* h2_outputs = apply(sigmoid, h2_add);
|
|
|
|
Matrix* h3_dot = dot(network->weights_3, h2_outputs);
|
|
Matrix* h3_add = add(h3_dot, network->bias_3);
|
|
Matrix* h3_outputs = apply(sigmoid, h3_add);
|
|
|
|
Matrix* final_dot = dot(network->weights_output, h3_outputs);
|
|
Matrix* final_add = add(final_dot, network->bias_output);
|
|
Matrix* final_outputs = apply(sigmoid, final_add);
|
|
|
|
Matrix* result = softmax(final_outputs);
|
|
|
|
matrix_free(h1_dot);
|
|
matrix_free(h1_add);
|
|
matrix_free(h1_outputs);
|
|
|
|
matrix_free(h2_dot);
|
|
matrix_free(h2_add);
|
|
matrix_free(h2_outputs);
|
|
|
|
matrix_free(h3_dot);
|
|
matrix_free(h3_add);
|
|
matrix_free(h3_outputs);
|
|
|
|
matrix_free(final_dot);
|
|
matrix_free(final_add);
|
|
matrix_free(final_outputs);
|
|
|
|
return result;
|
|
}
|
|
|
|
double cost_function(Matrix* calculated, int expected){
|
|
calculated->numbers[expected] -= 1;
|
|
apply(square, calculated);
|
|
|
|
// double loss = 0.5 * (target - output) * (target - output);
|
|
|
|
return 0;
|
|
}
|
|
|
|
void train_network(Neural_Network* network, Image *image, int label) {
|
|
|
|
// Flatten the image into matrix
|
|
Matrix* input = matrix_flatten(image->pixel_values, 0);
|
|
|
|
// Perform forward propagation
|
|
Matrix* h1_dot = dot(network->weights_1, input);
|
|
Matrix* h1_add = add(h1_dot, network->bias_1);
|
|
Matrix* h1_outputs = apply(sigmoid, h1_add);
|
|
|
|
Matrix* h2_dot = dot(network->weights_2, h1_outputs);
|
|
Matrix* h2_add = add(h2_dot, network->bias_2);
|
|
Matrix* h2_outputs = apply(sigmoid, h2_add);
|
|
|
|
Matrix* h3_dot = dot(network->weights_3, h2_outputs);
|
|
Matrix* h3_add = add(h3_dot, network->bias_3);
|
|
Matrix* h3_outputs = apply(sigmoid, h3_add);
|
|
|
|
Matrix* final_dot = dot(network->weights_output, h3_outputs);
|
|
Matrix* final_add = add(final_dot, network->bias_output);
|
|
Matrix* final_outputs = apply(sigmoid, final_add);
|
|
|
|
// begin backpropagation
|
|
Matrix* sigma = matrix_create(final_outputs->rows, 1);
|
|
matrix_fill(sigma, 1);
|
|
Matrix* temp1 = subtract(sigma, final_outputs);
|
|
Matrix* temp2 = multiply(temp1, final_outputs); // * soll-ist
|
|
Matrix* temp3 = matrix_create(final_outputs->rows, final_outputs->columns);
|
|
matrix_fill(temp3, 0);
|
|
temp3->numbers[label][0] = 1;
|
|
Matrix* temp4 = subtract(temp3, final_outputs);
|
|
sigma = multiply(temp2, temp4);
|
|
|
|
Matrix* temp5 = transpose(h3_outputs);
|
|
Matrix* temp6 = dot(sigma, temp5);
|
|
Matrix* weights_delta = scale(temp6, network->learning_rate);
|
|
Matrix* bias_delta = scale(sigma, network->learning_rate);
|
|
|
|
Matrix* temp7 = add(weights_delta, network->weights_output);
|
|
matrix_free(network->weights_output);
|
|
network->weights_output = temp7;
|
|
|
|
Matrix* temp8 = add(bias_delta, network->bias_output);
|
|
matrix_free(network->bias_output);
|
|
network->bias_output = temp8;
|
|
|
|
// other levels
|
|
backPropagation(network->learning_rate, network->weights_3, network->bias_3, h3_outputs, h2_outputs, sigma);
|
|
backPropagation(network->learning_rate, network->weights_2, network->bias_2, h2_outputs, h1_outputs, sigma);
|
|
backPropagation(network->learning_rate, network->weights_1, network->bias_1, h1_outputs, input, sigma);
|
|
|
|
matrix_free(input);
|
|
|
|
matrix_free(h1_dot);
|
|
matrix_free(h1_add);
|
|
matrix_free(h1_outputs);
|
|
|
|
matrix_free(h2_dot);
|
|
matrix_free(h2_add);
|
|
matrix_free(h2_outputs);
|
|
|
|
matrix_free(h3_dot);
|
|
matrix_free(h3_add);
|
|
matrix_free(h3_outputs);
|
|
|
|
matrix_free(final_dot);
|
|
matrix_free(final_add);
|
|
matrix_free(final_outputs);
|
|
|
|
matrix_free(weights_delta);
|
|
matrix_free(bias_delta);
|
|
|
|
matrix_free(sigma);
|
|
|
|
matrix_free(temp1);
|
|
matrix_free(temp2);
|
|
matrix_free(temp3);
|
|
matrix_free(temp4);
|
|
matrix_free(temp5);
|
|
matrix_free(temp6);
|
|
matrix_free(temp7);
|
|
matrix_free(temp8);
|
|
}
|
|
|
|
void backPropagation(double learning_rate, Matrix* weights, Matrix* biases, Matrix* current_layer_activation, Matrix* previous_layer_activation, Matrix* sigma_old) {
|
|
Matrix* sigma_new = matrix_create(current_layer_activation->rows, 1);
|
|
matrix_fill(sigma_new, 1);
|
|
|
|
Matrix* temp1 = subtract(sigma_new, current_layer_activation);
|
|
Matrix* temp2 = multiply(temp1, current_layer_activation); // *sum(delta*weights)
|
|
|
|
for(int i = 0; i < current_layer_activation->rows; i++) {
|
|
double sum = 0;
|
|
for (int j = 0; j < sigma_old->rows; j++) {
|
|
sum += current_layer_activation->numbers[i][j] * sigma_old->numbers[j][0];
|
|
}
|
|
temp1->numbers[i][0] = sum;
|
|
}
|
|
sigma_new = multiply(temp2, temp1);
|
|
|
|
// new sigma done
|
|
|
|
Matrix* temp3 = transpose(previous_layer_activation);
|
|
Matrix* temp4 = dot(sigma_new, temp3);
|
|
Matrix* weights_delta = scale(temp4, learning_rate);
|
|
Matrix* bias_delta = scale(sigma_new, learning_rate);
|
|
|
|
Matrix* temp5 = add(weights_delta, weights);
|
|
free(weights->numbers);
|
|
weights->numbers = temp5->numbers;
|
|
|
|
Matrix* temp6 = add(bias_delta, biases);
|
|
free(biases->numbers);
|
|
biases->numbers = temp6->numbers;
|
|
|
|
sigma_old->rows = sigma_new->rows;
|
|
sigma_old->columns = sigma_new->columns;
|
|
free(sigma_old->numbers);
|
|
sigma_old->numbers = sigma_new->numbers;
|
|
|
|
free(sigma_new);
|
|
|
|
matrix_free(temp1);
|
|
matrix_free(temp2);
|
|
matrix_free(temp3);
|
|
matrix_free(temp4);
|
|
matrix_free(temp5);
|
|
matrix_free(temp6);
|
|
matrix_free(weights_delta);
|
|
matrix_free(bias_delta);
|
|
}
|
|
|
|
|
|
//void batch_train_network(Neural_Network* network, Image** images, int size);
|
|
|
|
double sigmoid(double input) {
|
|
return 1.0 / (1 + exp(-1 * input));
|
|
}
|
|
|
|
double sigmoid_derivative(double x) {
|
|
return x * (1.0 - x);
|
|
}
|
|
|
|
Matrix* softmax(Matrix* matrix) {
|
|
double total = 0;
|
|
|
|
for (int i = 0; i < matrix->rows; i++) {
|
|
for (int j = 0; j < matrix->columns; j++) {
|
|
total += exp(matrix->numbers[i][j]);
|
|
}
|
|
}
|
|
Matrix* result_matrix = matrix_create(matrix->rows, matrix->columns);
|
|
for (int i = 0; i < result_matrix->rows; i++) {
|
|
for (int j = 0; j < result_matrix->columns; j++) {
|
|
result_matrix->numbers[i][j] = exp(matrix->numbers[i][j]) / total;
|
|
}
|
|
}
|
|
return result_matrix;
|
|
}
|
|
|
|
double square(double input) {
|
|
return input * input;
|
|
}
|
|
|
|
double loss_function(Matrix* output_matrix, int image_label) {
|
|
Matrix* temp = matrix_copy(output_matrix);
|
|
|
|
temp->numbers[1, image_label] -= 1;
|
|
apply(square, temp);
|
|
|
|
matrix_free(temp);
|
|
|
|
return matrix_sum(temp);;
|
|
} |