eigen_test_drive.cpp

//g++ -std=c++17 -O3 -I eigen main.cpp -o main && time ./main
//g++ -std=c++17 -fopenmp -O3 -I eigen main.cpp -o main && time ./main
//g++ -std=c++17 -pg -fopenmp -O3 -I eigen main.cpp -o main && time ./main && gprof main gmon.out > profile.txt
#include <iostream>
#include <Eigen/Dense>
#include <cmath>

using namespace Eigen;

float logistic(float x) {
  return 1.0 / (1.0 + std::exp(-x));
}

void nn_test() {
  //Eigen::setNbThreads(10);
  //std::cout << "Eigen is using " << Eigen::nbThreads() << " threads." << std::endl;

  int n = 252*100;

  std::cout << "\nforward application of neural network: " << n << " trials" << std::endl;
  auto layer1 = MatrixXf::Random(128, 588); // 588 is the number of columns in PDI variables file
  auto layer2 = MatrixXf::Random(32, 128);
  auto layer_out = MatrixXf::Random(1, 32);
  auto input = Matrix<float, 588, 1>::Random();

  for ( int i=1; i<=n; i++ ) {
    auto activation = (layer_out * (layer2 * (layer1 * input).unaryExpr(&logistic)).unaryExpr(&logistic)).unaryExpr(&logistic);

    if ( i > 1 and i % (n/10) == 0 or activation(0,0) == 0.2 ) {
      std::cout << i << "th score: " << activation(0,0) << std::endl;
    }
  }
}

void basic_tests() {
  //MatrixXd m(2,2);
  Matrix2f m(2,2);
  m(0,0) = 2;
  m(1,0) = 0.5;
  m(0,1) = m(1,0);
  m(1,1) = 1;

  std::cout << "Original:" << std::endl;
  std::cout << m << std::endl;
  std::cout << std::endl;

  Matrix2f mi = m.inverse();

  std::cout << "Inverse:" << std::endl;
  std::cout << mi << std::endl;
  std::cout << std::endl;

  ComplexEigenSolver<Matrix2f> e(m);
  if ( e.info() == Success ) {
    std::cout << "Eigenvalues:\n" << e.eigenvalues() << std::endl;
    std::cout << "\nEigenvector matrix:\n" << e.eigenvectors() << std::endl;
  } else {
    std::cout << "failed to solve the eigen-problem" << std::endl;
  }
}

int main(int argc, char** argv) {
  nn_test();
  return 0;
}