[Fwd: Re: Generalized Inverse?]

[Ron Boisvert <boisvert@nist.gov>]

Sender: Kim van der Linde <kim@kimvdlinde.com>
Subject: Re: Generalized Inverse?


Ok, it is clear, it did not work, I run into singular matrices 
regardless. Or, I probably need the generilized (Moore-Penrose) inverse 
anyway. Question, is this to be added to JAMA, or should I deal with it 
myself?

Cheers,

Kim

Kim van der Linde wrote:

> Hi,
> 
> I have a method that by definition produces a singular matrix, and I 
> need the inverse of that without loss of dimensions. It is a square 
> variance-covariance matrix that I use to estimate mahalanobis distances. 
> I think the Moore-Penrose pseudoinverse is the one I usually run into 
> when I search for solutions, but I am open to any other solution.
> 
> The next step is to use a Cholesky decomposition on the inverse which I 
> use then in the Weighted Generalized Procrustes Alignment Analysis. 
> After some digging today, I see I can maybe do the cholesky first and 
> take the inverse of the L and L-transposed. I am gong to try that.


-- 
http://www.kimvdlinde.com