Hi,
I am new to libeigen and have recieved the task to port some matlab code snippets to C++.
I have read the Core Tutorialbut have some questions about how to reimplement few Matlab functions for use with eigen (if they are not already available..)

repmat [Link to Doc]
B = repmat(A,m,n) creates a large matrix B consisting of an m-by-n tiling of copies of A. The size of B is [size(A,1)*m, (size(A,2)*n]. The statement repmat(A,n) creates an n-by-n tiling.
I think in pseudocode it should be reproducable by something like ones(m,n).*.A , but what is the best way to do it in eigen, using VectorXf as example.

meshgrid [Link to Doc]
Is there a simpel way to reproduce a matlab call like [X,Y] = meshgrid(1:3,10:14) without loops or something? The first problem I see is returning 2 matrices from a function, the second is I haven't found the matlab ":" operator, and I think I can only emulate it using a loop.

reshape [Link to Doc]
B = reshape(A,m,n) returns the m-by-n matrix B whose elements are taken column-wise from A. An error results if A does not have m*n elements.

Is any of this functionality already present in eigen and could I only not find it? Or has anyone suggestions on how to implement them in an elegant way?

regards,
Ismael

repmat

e.g., dst = src.replicate(n,m);

see http://eigen.tuxfamily.org/dox-devel/cl ... 4e4b103032

meshgrid

you can mix LinSpaced and replicate:

X = RowVectorXd::LinSpaced(1,3,3).replicate(5,1);
Y = VectorXd::LinSpaced(10,14,5).replicate(1,3);

see: http://eigen.tuxfamily.org/dox-devel/cl ... a9c6ae34d8

reshape

Currently you can mimic it using Map:

B = Map<MatrixXd>(A.data(),m,n);

... and do not miss this one
http://eigen.tuxfamily.org/dox-devel/As ... erence.txt
--Frank

Thanks for your fast help and explanations FMD and ggael
The QuickReference is realy great.

One thing however I couldn't find out myself.
I have to work most of the time with dynamic-sized vectors and need
to do something like this in matlab (simplified example):

Code: Select all

 a = [a; ones(10, 1)*17];

so I can use VectorXi b = VectorXi::Ones(10,1)*17 or VectorXi b = VectorXi::Constant(10,1,17); to compute the second part.
But how can I then combine them to one?
I could not find anything in the QuickReference as it mostly shows operations with matrices.

Do I have to manually resize a to the length of a+b, and then copy the values over in a loop? This seems to be a costly operation, and since I have to do it thousand of times I would like to implement it the right way.

thanks,
Ismael

You can do:

Code: Select all

result << a , VectorXi::Constant(10,1,17);


if the result is not the same matrix as a. However if it is a itself that you are extending, with eigen 3, you can do this:

Code: Select all

a.conservativeResize(a.size() + 10);
a.tail(10) = VectorXi::Constant(10,1,17);


ggael wrote:reshape

Currently you can mimic it using Map:

B = Map<MatrixXd>(A.data(),m,n);

Looks like this is affected by the matrix order. If A uses column-major, I don't think that your technique will produce the desired result. Can you please confirm??

someguy wrote:

ggael wrote:reshape

Currently you can mimic it using Map:

B = Map<MatrixXd>(A.data(),m,n);

Looks like this is affected by the matrix order. If A uses column-major, I don't think that your technique will produce the desired result. Can you please confirm??

I'm including source code to show this is the case:

Code: Select all

$ cat test_eigen_reshape.cpp
#include <Eigen/Core>
#include <Eigen/Array>

USING_PART_OF_NAMESPACE_EIGEN

typedef Eigen::Matrix<double, Eigen::Dynamic, Eigen::Dynamic> EigenMatrixDbl;


int main(int argc, char** argv)
{
        EigenMatrixDbl m(3, 4);
        m << 1, 2, 3, 3.5,
             4, 5, 6, 6.5,
             7, 8, 9, 9.5;

        std::cout << m << std::endl;

        const EigenMatrixDbl::RowXpr &first_row = m.row(0);

        std::cout << first_row << std::endl;

        Eigen::Map<EigenMatrixDbl> B = Eigen::Map<EigenMatrixDbl>(first_row.data(),2,2);

        std::cout << B << std::endl;

        return 0;
}


Code: Select all

$ g++ -o test_eigen -I./eigen test_eigen_reshape.cpp && ./test_eigen
1 2 3 3.5
4 5 6 6.5
7 8 9 9.5

1 2 3 3.5

1 7
4 2


This is wrong. The last result should be:

Code: Select all

1 2
3 3.5

Instead of:

Code: Select all

1 7
4 2

Any clue on how to get the right result??

someguy wrote:Any clue on how to get the right result??

In Eigen3 you can make use of the Stride object

Stride<Dynamic,Dynamic> stride(3,6);
const auto B = EigenMatrixDbl::Map(m.data(), 2, 2, stride);

but again, your result will depend on the memory layout. Right now we don't have a "real" reshape.

- Hauke

Hauke wrote:In Eigen3 you can make use of the Stride object

Stride<Dynamic,Dynamic> stride(3,6);
const auto B = EigenMatrixDbl::Map(m.data(), 2, 2, stride);

Thanks, but I use 2.0.15. Version 3 is still in Beta.

Hauke wrote:but again, your result will depend on the memory layout. Right now we don't have a "real" reshape.

This is what I came up with:

Code: Select all

#include <Eigen/Core>
#include <Eigen/Array>

USING_PART_OF_NAMESPACE_EIGEN

typedef Eigen::Matrix<double, Eigen::Dynamic, Eigen::Dynamic> EigenMatrixDbl;
typedef Eigen::Matrix<double, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor> EigenMatrixDblRowMajor;


int main(int argc, char** argv)
{
        EigenMatrixDbl m(3, 4);
        m << 1, 2, 3, 3.5,
             4, 5, 6, 6.5,
             7, 8, 9, 9.5;

        std::cout << m << std::endl << std::endl;

        const EigenMatrixDbl::RowXpr &first_row = m.row(0);
            
   std::cout << first_row << std::endl << std::endl;

   Eigen::Map<EigenMatrixDbl> B = Eigen::Map<EigenMatrixDbl>(first_row.data(),2,2);

   std::cout << B << std::endl << std::endl;

   EigenMatrixDblRowMajor n(1, m.cols());
   n << first_row;
            
   Eigen::Map<EigenMatrixDblRowMajor> C = Eigen::Map<EigenMatrixDblRowMajor>(n.data(),2,2);            
   std::cout << C << std::endl << std::endl;

        return 0;
}



Messy, but seems to work

More generically, for a matrix, the following code should give the correct reshape:

Code: Select all

#include <Eigen/Core>
#include <Eigen/Array>

USING_PART_OF_NAMESPACE_EIGEN

typedef Eigen::Matrix<double, Eigen::Dynamic, Eigen::Dynamic> EigenMatrixDbl;
typedef Eigen::Matrix<double, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor> EigenMatrixDblRowMajor;

int main(int argc, char** argv)
{
        EigenMatrixDbl m(3, 4);
        m << 1, 2, 3, 3.5,
             4, 5, 6, 6.5,
             7, 8, 9, 9.5;

        std::cout << m << std::endl << std::endl;


        EigenMatrixDblRowMajor n(3, 4);
        n << m;
        std::cout << n << std::endl << std::endl;

        Eigen::Map<EigenMatrixDblRowMajor> B2 = Eigen::Map<EigenMatrixDblRowMajor>(n.data(),4,3);

        std::cout << B2 << std::endl << std::endl;

        return 0;
}


Hi,

I too am looking for a conversion from Matlab to Eigen on repmat, specifically this line.

X = X ./ repmat( sum(X, 1), size(X, 1), 1 );

My initial thoughts were to use a cwiseQuotient and a replicate with the relevant eigen reduction calls but this doesn't seem to be working.

Any ideas would be great!

Thanks,
Kris

What about the following (not tested:

Code: Select all

X = X.array().rowise() / X.colwise().sum().array();