Re: [eigen] Eigen "views"

[Gael Guennebaud <gael.guennebaud@xxxxxxxxx>]

I confirm what Benoit said, this has yet to be implemented and this is
probably the most wanted feature!

API-wise, what about simply overloading operator(). For vectors:

VectorXf v1, v2;
VectorXi indices;

v2 = v1(indices);

And for matrices:

MatrixXf m1, m2;
VectorXi indices1, indices2;

m2 = m1(indices1,indices2);
m2 = m1(Eigen::All,indices2);
m2 = m1(indices1,Eigen::All);

"m1(indices1)" would be forbidden for matrices.

Maybe the constant "All" is too intrusive when someone do using
namespace Eigen ??


Gael


On Mon, Aug 15, 2011 at 9:24 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
> (Disclaimer: I've been away for a while so I might not be up to date.
> If Gael contradicts me, listen to him)
>
> This has been asked for many times, and I think there was consensus
> that this would be a welcome addition. It also isn't hard to
> implement. It's just that nobody has actually done the work yet :)
>
> If you want to do it yourself, the right Eigen way to do it is to add
> a new expression class, holding references to the original vector and
> to the integer vector of indices. For a simple expression class to
> start from as an example, look at the Minor class in Eigen2Support,
> it's as easy as it gets. The core part of the implementation would be
> the coeff() accessor methods in this class, and they would look like:
>
>   CoeffReturnType MyViewExpression::coeff(Index i)
>   {
>      return m_originalVector.coeff(m_indicesVector.coeff(i));
>   }
>
> I'll be happy to provide guidance.
>
> Cheers,
> Benoit
>
> 2011/8/15 Radu B. Rusu <rusu@xxxxxxxxxxxxxxxx>:
>> Hi guys,
>>
>> We were wondering whether there was ever a discussion about implementing
>> "views" into Eigen. Here's an example:
>>
>> Eigen::Vector<dynamic, float> vector1, vector2;
>> vector1 = defineSomeDataSomehow ();
>> Eigen::Vector<dynamic, int> idx = defineSomeIndicesSomehow ();
>> vector2 = vector1 (idx);
>> assert (vector2.size () == idx.size ());
>>
>>
>> Basically addressing data without copies.
>>
>> Thanks,
>> Radu.
>> --
>> Point Cloud Library (PCL) - http://pointclouds.org
>>
>>
>>
>
>
>