I am studying the low rank approximations of matrices, for example by truncated the Singular Value Decomposition of a matrix.
In order to test any algorithm, I need some test matrices that have internal structure, for example coming from modelisation of a physical problem (e.g. diffusion problem). Indeed, I have experienced that these matrices tend to be good candidates (or very bad) for matrix approximation.
Is there a project out there that enables to generate matrices from real life problems ?
Thank you for any information.