#include <iostream>
#include <cmath>
#include <ctime>
#include <cstdlib>
const int rows = 4, cols = 3;
double sigmoid(double x, double w)
{
	if (x == 0) x -= w;
	//if (x > -1 && x<1) x -= w;
	return 1 / (1 + exp(-(x)));
}
double fRand()
{
	double fMin = -1;
	double fMax = 1;
	double f = (double)rand() / RAND_MAX;
	return fMin + f * (fMax - fMin);
}
template <int Size1, int Size2>
double newDot(int(&input)[Size1][Size2], double other[], int k)
{
	double d = 0;
	for (int j = 0; j < Size2; j++)
	{
		d += input[k][j] * other[j];
	}
	return d;
}
void transpose(int T[cols][rows], int input_layers[rows][cols])
{
	for (int i = 0; i < rows; i++)
		for (int j = 0; j < cols; j++)
			T[j][i] = input_layers[i][j];
}
void findRfromO(double errors[rows], double outputs[rows], double right[rows])
{
	for (int i = 0; i < rows; i++)
	{
		right[i] = errors[i] * (outputs[i] * (1 - outputs[i]));
	}
}
int main()
{
	double B = 8;
	srand(time(0));
	int training_inputs[rows][cols] =
	{ {0,1,0},
	{0,1,1},
	{0,0,1},
	{1,0,1} };
	int training_outputs[rows] = { 0,0,0,1 };
	double synaptic_weights[cols];
	for (int i = 0; i < cols; i++)
		synaptic_weights[i] = fRand();
	std::cout << "Starting weights: " << std::endl;
	for (int i = 0; i < cols; i++)
		std::cout << synaptic_weights[i] << std::endl;
	//Метод обратного распространения
	int input_layers[rows][cols];
	for (int i = 0; i < rows; i++)
		for (int j = 0; j < cols; j++)
			input_layers[i][j] = training_inputs[i][j];
	double outputs[rows];
	double errors[rows];
	double adjustments[cols];
	double right[rows];
	int T_input_layers[cols][rows];
	for (int g = 0; g < 20000; g++)
	{
		for (int i = 0; i < rows; i++)
			outputs[i] = sigmoid(newDot(input_layers, synaptic_weights, i), B);
		for (int i = 0; i < rows; i++)
			errors[i] = training_outputs[i] - outputs[i];
		transpose(T_input_layers, input_layers);
		findRfromO(errors, outputs, right);
		for (int j = 0; j < cols; j++)
			adjustments[j] = newDot(T_input_layers, right, j);
		for (int j = 0; j < cols; j++)
			synaptic_weights[j] += adjustments[j];
	}

	std::cout << "\nWeights after learning: " << std::endl;
	for (int j = 0; j < cols; j++)
		std::cout << synaptic_weights[j] << std::endl;

	std::cout << "\nResult for starting inputs: " << std::endl;
	for (int i = 0; i < rows; i++)
		std::cout << outputs[i] << std::endl;


	//Тест
	const int new_rows = 2;
	int new_in[new_rows][cols];
	int x;
	for (int i = 0; i < new_rows; i++)
	{
		for (int j = 0; j < cols; j++)
		{
			std::cin >> x;
			new_in[i][j] = x;
		}
	}
	double new_out[new_rows];
	std::cout << "\n\nTest Results: " << std::endl;
	for (int i = 0; i < new_rows; i++)
	{
		new_out[i] = sigmoid(newDot(new_in, synaptic_weights, i), B);
		std::cout << std::endl << new_out[i];
	}
}