Eigen equivalent of Matlab rescale command

Question

I'd like to recreate the Matlab rescale command in Eigen

https://www.mathworks.com/help/matlab/ref/rescale.html

I've tried to translate but not sure about it. My Eigen knowledge is too limited still...

  auto rescale = [&](
    Eigen::MatrixXd mat, 
    Eigen::VectorXd inmin,
    Eigen::VectorXd inmax,
    Eigen::VectorXd l,
    Eigen::VectorXd u
    ) -> Eigen::MatrixXd {

    auto val = l.array() + (
      ((mat - inmin).array()) / ((
        ((inmax - inmin).array()) * ((u - l).array())
      ).array())
    );

    return val;
  };

could this be viable?

Sure it could be. Define your requirements for this function and write the tests to check if it behaves as you wish. In matlab docs the requirements for `rescale` are not that trivial. — pptaszni, May 13 '19 at 16:27
The matlab rescale command is here https://www.mathworks.com/matlabcentral/answers/196798-how-to-normalize-values-in-a-matrix-to-be-between-0-and-1 — Kabu, May 14 '19 at 19:37

Avi Ginsburg · Accepted Answer · 2019-05-13T18:29:12.813

No. Your dimensions don't match. You're mixing ArrayXd and ArrayXXd. This is more like what you want, with a version for scalars and another for vectors. Adjust the rowwise/colwise to match the different versions of matlabs rescale.

#include <Eigen/Core>
#include <iostream>

int main()
{

    auto rescale = [&](
        Eigen::MatrixXd mat,
        double l,
        double u
        ) -> Eigen::MatrixXd {

        double min = mat.minCoeff();
        double max = mat.maxCoeff();
        auto val = l + (mat.array() - min) * ((u - l) / (max - min));
        return val;
    };

    Eigen::MatrixXd mat(4,4);
    Eigen::Map<Eigen::VectorXd>(mat.data(), mat.size()).setLinSpaced(1, mat.size());

    std::cout << mat << "\n\n";

    auto rescaled = rescale(mat, 2, 5);

    std::cout << rescaled << "\n\n";

    auto rescale2 = [&](
        Eigen::MatrixXd mat,
        Eigen::VectorXd l,
        Eigen::VectorXd u
        ) -> Eigen::MatrixXd {

        Eigen::ArrayXd  min = mat.colwise().minCoeff();
        Eigen::ArrayXd  max = mat.colwise().maxCoeff();
        auto val = l.array().replicate(1, mat.cols())
            + ((mat.array().rowwise() - min.transpose()).rowwise() * 
               ((u - l).array() / (max - min)).transpose());
        return val;
    };

    Eigen::VectorXd mn, mx;

    mn.resize(mat.cols());
    mx.resize(mat.cols());

    mn.setConstant(1.3);
    mx << 2, 5, 6, 9;
    rescaled = rescale2(mat, mn, mx);

    std::cout << rescaled << "\n\n";



    return 0;
}

Eigen equivalent of Matlab rescale command

1 Answers1

Linked