double derv(double *w) {
*w += 0.0000001;
double x = pow(feed_forward() - f(input), 2); // E(x+0.00001)
*w -= 0.000002;
double y = pow(feed_forward() - f(input), 2); // E(x-0.00001)
*w += 0.0000001;
double diff = x - y;
return diff / 0.0000001;
}
void MultilayerPerceptron::train_sync(cv::Mat train_data,
cv::Mat expected_outputs)
{
cv::Mat output;
for(int i = 0; i < train_data.rows; ++i)
{
output = feed_forward(train_data.row(i));
backpropagation(expected_outputs.row(i), output);
update_weights();
}
}
void train(void) {
int i, j, k, l;
for (i = 0; i < sizeof trset / sizeof trset[0]; i++) {
printf("\nTrainning Round %d\n", i);
// feed an input into the network
input = trset[i];
feed_forward();
printe();
// back-propagate
Neuron cpy[LAYER_NUM][NEURON_PER_LAYER];
memcpy(cpy, hlayers, sizeof cpy);
for (j = 0; j < LAYER_NUM; j++) {
for (k = 0; k < NEURON_PER_LAYER; k++) {
if (j == 0) {
cpy[0][k].weights[0] = hlayers[0][k].weights[0] -
derv(&hlayers[0][k].weights[0]) * RATE;
}else {
for (l = 0; l < NEURON_PER_LAYER; l++) {
cpy[j][k].weights[l] = hlayers[j][k].weights[l] -
derv(&hlayers[j][k].weights[l]) * RATE;
}
}
}
}
Neuron outcpy = output;
for (j = 0; j < NEURON_PER_LAYER; j++) {
outcpy.weights[j] = output.weights[j] -
derv(&output.weights[j]) * RATE;
}
memcpy(hlayers, cpy, sizeof hlayers); // update weights
output = outcpy;
// printw();
}
}
