|Re: [eigen] Does anyone know of a good l1-solver?|
[ Thread Index | Date Index | More lists.tuxfamily.org/eigen Archives ]
> Either that, or maybe find x, s.t. ||Ax - b||_1=min ?
Yes this is what I want to solve although I want to add a l2 regulariser sometime later to this.
> Do you know anything about A? (sparse, square, symmetric, rank-deficient, ill-conditioned, ...)
No. A is a dense matrix and full rank but I do know that it has 6 columns and 3N rows.
> Anyways, afaik, Eigen itself does not provide L1-solvers, so you need to find an external library anyways (or find an algorithm and implement it). It is quite easy to interact with other libraries by using Eigen's Map and .data() functionality.
Hmmm... I thought it would be nice if there was a efficient library which did this since I run the optimisation in a loop.
Also is the primal dual technique the most efficient for this problem? That's what I found in a matlab library called l1 magic.
|Mail converted by MHonArc 2.6.19+||http://listengine.tuxfamily.org/|