Re: [eigen] SVD with singular matrices

[Cyrille Berger <cberger@xxxxxxxxxxx>]

On torsdag 30 september 2010, Benoit Jacob wrote:

> 2010/9/30 <hamelin.philippe@xxxxxxx>:

> > Knowing there is a major problem with the current SVD, I think it

> > shouldn't be available in the API until the problem is resolved. So both

> > options are valid and the choice mainly relies on that question: in the

> > long term, do you want to maintain 2 SVD algorithms ?

>

> Yes, I do! Because any SVD implementation is a trade-off between

> reliability and speed. There are two major families of algorithm:

>

> 1. Jacobi methods (reliable and typically slow)

> 2. Bidiagonal methods (not so reliable and typically faster)

Why not call the second one FastSVD (when fixed) ? Then people will have an incitive to look at the documentation before using it.

--

Cyrille Berger