Hello,

I discovered Eigen a few days ago, and it seems to be a great matrix library. So I'm trying to use Eigen instead of an old library developped internaly. The problem is that i want to preserve the compatibilty with the programs written before.
I read the page http://eigen.tuxfamily.org/dox/CustomizingEigen.html but I don't understand how I can create fonctions using Eigen.
For instance I had a fonction Ones to create a matrix of 1. the prototype was friend matrix & Ones(int nr, int nc), and I can use it with
matrix A; A = Ones(5,5);
So I'm looking for a pratical solution

Thank you in advance for any help !

For this particular function, it turns out that we have exactly the same in Eigen, as static method:

Code: Select all

MatrixXf a = MatrixXf::Ones(5,5);


Writing such a function in an efficient way is nontrivial, since for efficiency you want it to return an expression object, not a plain matrix. See Eigen's source code for the definition of Ones returning a CwiseNullaryOp.

Thanks for your quick answer

But I'm looking for a more general solution for writing functions that return a matrix. For instance I had a function that delete a row, function that return matrix with a gaussian (or gamma, beta, etc. ) distribution. And I don't know how I can write such a fonction in an "addon" file extending the MatrixBase class :S

I can write a new class with a MatrixXd as parameter but it will be unefficient...

I think I have found how i can do that with functions outside the class !

But I have one more question : is it possible to enable a "cwise" mode that is to say can cwise be implicit to make for instance m <= n instead of m.cwise() <= n ?

Thanks !

If you want cwise to be implicit, you want eigen3 (devel branch)'s Array class.

There is nothing like that in eigen2, unfortunately.

for that you need to implement all the operators you want <= in the MatrixBase addons file.

In the devel branch (Eigen3, first beta on Wednesday), we have Array objects which are just like Matrix and vectors but specialized for coeff-wise operations. To see the incompatibilities between Eigen 2 and 3 see this page:

http://eigen.tuxfamily.org/dox-devel/Ei ... igen3.html

But I'm not sure I can use the third version, because my program needs to be stable :S Perhaps I need to wait for the beta ?

I try to put this code in my addon file but it doesn't work (I had no hope but I tried ^^), the problem is that the operator <= is already surcharged.

Code: Select all

matrix operator<=(const matrix &m,const matrix &n)
{
   matrix z = (m.cwise() <= n);
   return z;
}

hm, you should not return a "matrix", what is matrix ? no, you should return an expression, just like .cwise()<= does. Try this:

Code: Select all

template<typename OtherDerived>
inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)
operator<=(const MatrixBase<OtherDerived> &other) const
{
  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)(derived(), other.derived());
}

this should really work.

Yes, sorry matrix was something specific to my program (to preserve the compatibility) : #define matrix MatrixXd

I tried our code in a EIGEN_CWISE_PLUGIN but I have an error :
error C2535: 'const Eigen::CwiseBinaryOp<std::less_equal<ei_traits<T>::Scalar>,ExpressionType,OtherDerived> Eigen::Cwise<Derived>::operator <=(const Eigen::MatrixBase<OtherDerived> &) const' : fonction membre déjà définie ou déclarée (function already declared)
src/Core/Cwise.h(141) : voir la déclaration de 'Eigen::Cwise<Derived>::operator <='

replace

EIGEN_CWISE_PLUGIN

by

EIGEN_MATRIXBASE_PLUGIN

Also you should not have to declare these functions using an explicit matrix type to have a source compatibility. With the above addons, the following will compile fine:

matrix a, b;
a<=b


Also, Re-reading your example, it should also work if you declare it in one of your header files, but *not* as a Eigen addons. Eigen's addons are inserted into their respective class definition. Inside Eigen, we have:

Code: Select all

class MatrixBase {
//...
#ifdef EIGEN_MATRIXBASE_PLUGIN
#include EIGEN_MATRIXBASE_PLUGIN
#endif

};

Thanks Yes I knew that the addons are really include in the class. (Actually it's pratical and good to know)
But for my previous problem I have still a problem :S

I have place the code

Code: Select all

(line 115) template<typename OtherDerived>
inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)
operator<=(const MatrixBase<OtherDerived> &other) const
{
  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)(derived(), other.derived());
}

in the EIGEN_MATRIXBASE_PLUGIN but I have already an error : (sorry it's in french, i can translate if someone doesn't understand)

error C4430: spécificateur de type manquant - int est pris en compte par défaut. Remarque : C++ ne prend pas en charge int par défaut
1> c:\documents and settings\bureau\eigen\eigen\src/Core/MatrixBase.h(643) : voir la référence à l'instanciation de la classe modèle 'Eigen::MatrixBase<Derived>' en cours de compilation
1>c:\documents and settings\bureau\eigen\eigen\testeigen\extension.h(117) : error C2143: erreur de syntaxe : absence de ';' avant '<='
.... etc. .... (45 errors....)

And if I use my code

Code: Select all

matrix operator<=(const matrix &m,const matrix &n)
{
   matrix z = (m.cwise() <= n);
   return z;
}

in an "normal" header file I have the error :

error C2039: 'YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY' : n'est pas membre de 'Eigen::ei_static_assert<condition>'

I don't understand this error

The second error that you're getting, YOU_MIXED_DIFFERENT_NUMERIC_TYPES, is because when you do:

matrix z = (m.cwise() <= n);

You are assigning the *boolean* matrix expression (m.cwise() <= n) to the matrix z. The best fix is to change your operator to return a boolean matrix. (See how Eigen does). If however you prefer to return the same matrix type, you need to cast to your scalar type:

matrix z = (m.cwise() <= n).template cast<Scalar>();

Thank you It works !

(I will use this method until Eigen 3)

Hi,

I have a new question I'm trying to add a new function in order to do new affectations to a matrix. For example :
A = 1.0; and all coefficients of A will be 1.

So I write a fonction that seems to be ok but when I want to use it, this function is not used and I have an error...

Code: Select all

EIGEN_STRONG_INLINE Derived& operator=(const Scalar &v)
{
   for (int i = 0; i < rows(); i++)
      for(int j = 0; i < cols(); j++)
         derived()(i,j) = v;
   return (*this);
}


The error :

error C2664: 'Eigen::Matrix<_Scalar,_Rows,_Cols> &Eigen::Matrix<_Scalar,_Rows,_Cols>::operator =(const Eigen::Matrix<_Scalar,_Rows,_Cols> &)' : impossible to convert the first parameter 'double' in 'const Eigen::Matrix<_Scalar,_Rows,_Cols> &'

If you have an idea

Here 1.0 is a double constant, so, since you get this error, I guess the Scalar type of your matrix is not double. If you have a matrix of floats, try A = 1.0f, your operator should be used.

Also, look again at your second for statement, you have a bug in there (i versus j).