//
// Created by Denis Ryapolov on 04.03.2024.
//
#include "sin_test.h"

void sin_test::test() {
  std::random_device rd;
  std::mt19937 gen(42);
  std::uniform_real_distribution<> dis(
      0, M_PI + 0.01);
  int size = 500;
  vector<double> in_values(size);
  vector<double> target_values(size);
  for (int i = 0; i < size; ++i) {
    double randomNum = dis(gen);
    in_values[i] = randomNum;
    target_values[i] = sin(randomNum);
  }
  double min_score = 1e9;
  double optimal_step = -1;
  for (int step_size = 1; step_size < 150; ++step_size){
    Network net({1, 50, 50, 1}, {ThresholdId::Sigmoid, ThresholdId::Sigmoid,
                                 ThresholdId::Sigmoid});
    if (step_size % 100 == 0){
      std::cout << step_size << "\n";
    }
    for (int epoch = 0; epoch < 10; ++epoch) {
      //      std::cout << "Epoch num: " << epoch << '\n';
      //      std::cout << '\n';
      for (int k = 0; k < size; ++k) {
        MatrixXd input(1, 1);
        MatrixXd target(1, 1);
        input << in_values[k];
        target << target_values[k];
        net.Train(input, target, ScoreFunc::create(ScoreId::MSE), 1, 1.24);
      }
    }
    double score = 0;
    for (int i = 0; i < size; ++i) {
      double randomNum = dis(gen);
      VectorXd test_input(1, 1);
      test_input << randomNum;
      score += abs(net.Calculate(test_input)(0, 0) - sin(randomNum));
      if (step_size == 124){
        std::cout << net.Calculate(test_input) << " " << sin(randomNum) << '\n';
      }
      //    test_input << in_values[i];
      //    std::cout << net.Calculate(test_input) << " " << target_values[i] <<
      //    '\n';
    }
    if (score/size < min_score){
      optimal_step = double(step_size)/100;
      min_score = score/size;
    }
    //    std::cout << "step size: " << double(step_size)/100 << " score: "<< score / size << '\n';
  }
  std::cout << min_score << " " << optimal_step << '\n';
}