Hi,

I've spent the last two days comparing matrices libraries and I decided myself for Eigen, which sounds pretty amazing. I have to work with 3D and 4D datasets. I didn't succeed to get a matrix with dimensions higher than 2, and when I try to make a vector of matrices (or matrix of matrices), I get an error as it is not a valid Scalar.

Is there a way to do it ?
Is Eigen still a good choice for such kind of datasets (4 dimensions: time(>1000 values), height(1-3 values), longitudes(5-100 values), latitudes(5-100 values)) ?
I need high performance as a priority. My procedure is the following: I fill once the matrix and then iterate on the time axis to make some linear algebra between this "slice" and every other time steps. It means a lot of iterations on one matrix.

Thanks, Pascal

So you're actually searching for a Tensor library, not a matrix library?
If you want to iterate over matrices you can use a STL container like vector<>, see http://eigen.tuxfamily.org/dox/StlContainers.html

Yes, you're right, I work on tensors. However, the calculations are always processed at the matrix level. So the iteration over matrices should be a solution. Thanks for the link.