82 lines
2.9 KiB
C++
82 lines
2.9 KiB
C++
#include "core.h"
|
|
|
|
NeuralNetwork::NeuralNetwork(LayerStructure_t layers[], int count) {
|
|
numLayers = count;
|
|
for (int i = 0; i < count; i++) {
|
|
layerSizes.push_back(layers[i].size);
|
|
}
|
|
|
|
// Инициализация весов случайными числами
|
|
for (int i = 0; i < count - 1; i++) {
|
|
std::vector<std::vector<double>> layerWeights;
|
|
for (int j = 0; j < layerSizes[i+1]; j++) {
|
|
std::vector<double> nodeWeights;
|
|
for (int k = 0; k < layerSizes[i]; k++) {
|
|
nodeWeights.push_back(((double)rand() / RAND_MAX) * 2 - 1);
|
|
}
|
|
layerWeights.push_back(nodeWeights);
|
|
}
|
|
weights.push_back(layerWeights);
|
|
|
|
std::vector<double> layerBiases;
|
|
for (int j = 0; j < layerSizes[i+1]; j++) {
|
|
layerBiases.push_back(((double)rand() / RAND_MAX) * 2 - 1);
|
|
}
|
|
biases.push_back(layerBiases);
|
|
}
|
|
}
|
|
|
|
std::vector<double> NeuralNetwork::feedForward(std::vector<double> input) {
|
|
outputs.clear();
|
|
outputs.push_back(input);
|
|
|
|
std::vector<double> current = input;
|
|
for (int i = 0; i < numLayers - 1; i++) {
|
|
std::vector<double> next;
|
|
for (int j = 0; j < layerSizes[i+1]; j++) {
|
|
double sum = biases[i][j];
|
|
for (int k = 0; k < layerSizes[i]; k++) {
|
|
sum += current[k] * weights[i][j][k];
|
|
}
|
|
next.push_back(sigmoid(sum));
|
|
}
|
|
current = next;
|
|
outputs.push_back(current);
|
|
}
|
|
return current;
|
|
}
|
|
|
|
void NeuralNetwork::train(std::vector<double> input, std::vector<double> target, double lr) {
|
|
// 1. Прямой проход
|
|
feedForward(input);
|
|
|
|
// 2. Вычисление ошибок для выходного слоя
|
|
std::vector<std::vector<double>> errors(numLayers);
|
|
errors[numLayers - 1].resize(layerSizes[numLayers - 1]);
|
|
for (int i = 0; i < layerSizes[numLayers - 1]; i++) {
|
|
double output = outputs[numLayers - 1][i];
|
|
errors[numLayers - 1][i] = (target[i] - output) * sigmoidDerivative(output);
|
|
}
|
|
|
|
// 3. Обратное распространение ошибки на скрытые слои
|
|
for (int i = numLayers - 2; i > 0; i--) {
|
|
errors[i].resize(layerSizes[i]);
|
|
for (int j = 0; j < layerSizes[i]; j++) {
|
|
double error = 0.0;
|
|
for (int k = 0; k < layerSizes[i+1]; k++) {
|
|
error += errors[i+1][k] * weights[i][k][j];
|
|
}
|
|
errors[i][j] = error * sigmoidDerivative(outputs[i][j]);
|
|
}
|
|
}
|
|
|
|
// 4. Обновление весов и смещений
|
|
for (int i = 0; i < numLayers - 1; i++) {
|
|
for (int j = 0; j < layerSizes[i+1]; j++) {
|
|
for (int k = 0; k < layerSizes[i]; k++) {
|
|
weights[i][j][k] += lr * errors[i+1][j] * outputs[i][k];
|
|
}
|
|
biases[i][j] += lr * errors[i+1][j];
|
|
}
|
|
}
|
|
} |