rawalcher/c-net
Archived
1
0
Fork 0
Code Issues 5 Pull requests Projects Releases 2 Packages Wiki Activity Actions

small changes

Browse source
This commit is contained in:
Thomas Schleicher 2023-09-21 16:02:50 +02:00
parent ed563e1e9e
commit b2e59c9ad7
3 changed files with 37 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

8
main.c
View file

@ -5,13 +5,15 @@
#include "neuronal_network.h" #include "neuronal_network.h"
int main() { int main() {
// Image** images = import_images("../data/train-images.idx3-ubyte", "../data/train-labels.idx1-ubyte", NULL, 60000); Image** images = import_images("../data/train-images.idx3-ubyte", "../data/train-labels.idx1-ubyte", NULL, 60000);
// img_visualize(images[4]); // img_visualize(images[4]);
// Neural_Network* nn = new_network(4, 2, 3, 0.5); Neural_Network* nn = new_network(28*28, 16, 10, 0.5);
// randomize_network(nn, 20); randomize_network(nn, 20);
// save_network(nn); // save_network(nn);
// Neural_Network* nn = load_network("../networks/test1.txt"); // Neural_Network* nn = load_network("../networks/test1.txt");
train_network(nn, images[0], 5);
} }

46
neuronal_network.c
View file

@ -4,9 +4,12 @@
#include <time.h> #include <time.h>
#include <math.h> #include <math.h>
double relu(double input); double sigmoid(double input);
Matrix* softmax(Matrix* matrix); Matrix* sigmoidPrime(Matrix* m);
Matrix* softmax(Matrix* matrix);
double square(double input); double square(double input);
double loss_function(Matrix* output_matrix, int image_label); double loss_function(Matrix* output_matrix, int image_label);
Neural_Network* new_network(int input_size, int hidden_size, int output_size, double learning_rate){ Neural_Network* new_network(int input_size, int hidden_size, int output_size, double learning_rate){
@ -152,13 +155,13 @@ Matrix* predict_image(Neural_Network* network, Image* image){
} }
Matrix* predict(Neural_Network* network, Matrix* image_data) { Matrix* predict(Neural_Network* network, Matrix* image_data) {
Matrix* hidden1_outputs = apply(relu, add(dot(network->weights_1, image_data), network->bias_1)); Matrix* hidden1_outputs = apply(sigmoid, add(dot(network->weights_1, image_data), network->bias_1));
Matrix* hidden2_outputs = apply(relu, add(dot(network->weights_2, hidden1_outputs), network->bias_2)); Matrix* hidden2_outputs = apply(sigmoid, add(dot(network->weights_2, hidden1_outputs), network->bias_2));
Matrix* hidden3_outputs = apply(relu, add(dot(network->weights_3, hidden2_outputs), network->bias_3)); Matrix* hidden3_outputs = apply(sigmoid, add(dot(network->weights_3, hidden2_outputs), network->bias_3));
Matrix* final_outputs = apply(relu, add(dot(network->weights_output, hidden3_outputs), network->bias_output)); Matrix* final_outputs = apply(sigmoid, add(dot(network->weights_output, hidden3_outputs), network->bias_output));
Matrix* result = softmax(final_outputs); Matrix* result = softmax(final_outputs);
@ -176,18 +179,31 @@ double cost_function(Matrix* calculated, int expected){
} }
//void train_network(Neural_Network* network, Matrix* input, Matrix* output); void train_network(Neural_Network* network, Image *image, int label) {
//void batch_train_network(Neural_Network* network, Image** images, int size);
Matrix* input = matrix_flatten(image->pixel_values, 0);
Matrix* hidden1_outputs = apply(sigmoid, add(dot(network->weights_1, input), network->bias_1));
Matrix* hidden2_outputs = apply(sigmoid, add(dot(network->weights_2, hidden1_outputs), network->bias_2));
Matrix* hidden3_outputs = apply(sigmoid, add(dot(network->weights_3, hidden2_outputs), network->bias_3));
Matrix* final_outputs = apply(sigmoid, add(dot(network->weights_output, hidden3_outputs), network->bias_output));
double relu(double input) {
if (input <= 0){
return 0.0;
}
return input;
} }
double relu_derivative(double x) { //void batch_train_network(Neural_Network* network, Image** images, int size);
return (x > 0) ? 1 : 0;
double sigmoid(double input) {
return 1.0 / (1 + exp(-1 * input));
}
Matrix* sigmoidPrime(Matrix* m) {
Matrix* ones = matrix_create(m->rows, m->columns);
matrix_fill(ones, 1);
Matrix* subtracted = subtract(ones, m);
Matrix* multiplied = multiply(m, subtracted);
matrix_free(ones);
matrix_free(subtracted);
return multiplied;
} }
Matrix* softmax(Matrix* matrix) { Matrix* softmax(Matrix* matrix) {

2
neuronal_network.h
View file

@ -39,5 +39,5 @@ double measure_network_accuracy(Neural_Network* network, Image** images, int amo
Matrix* predict_image(Neural_Network* network, Image* image); Matrix* predict_image(Neural_Network* network, Image* image);
Matrix* predict(Neural_Network* network, Matrix* image_data); Matrix* predict(Neural_Network* network, Matrix* image_data);
void train_network(Neural_Network* network, Matrix* input, Matrix* output); void train_network(Neural_Network* network, Image *image, int label);
void batch_train_network(Neural_Network* network, Image** images, int size); void batch_train_network(Neural_Network* network, Image** images, int size);