Re: Octave/C++ matrix Inv() comparison

[ionone]

I computed the ratio of S as you told me and i end up with  this number : 1.9902e+000/2.6573e-014 = 7.49e13 so it's bigger than 1e8.
The thing is that i have good datas with Inv(), i mean they make sense. For example with datas growing from 1/1024 to 1 in steps of 1/1024 i have an e (in the Octave function i posted before) coherent : from example, the numbers follow  pattern : 0.024 0.0234 0.0231 0.0224, etc which i don't have using Lapack dgetri/dsytri

Jeff

Le 09/07/2013 15:46, CdeMills [via Octave] a écrit :

Hello,

before starting to invest your time into comparing algorithms and toolboxes ... first look at your data. Perform a singular value decomposition, look at the ratio of the biggest singular value (top left of the S matrix) to the smallest one (bottom right). A ratio greater than 1/sqrt(eps), around 1e8, is calling for rounding issues. If this is the case, try understanding why this ratio is so low: it is the symptom that some of your columns are strongly linearly correlated. Common cures: compute pseudo-inverse (pinv), i.e. neglecting "directions" where noise sensitivity is great; or minimal-norm solutions.

Regards

Pascal

---

If you reply to this email, your message will be added to the discussion below:

http://octave.1599824.n4.nabble.com/Octave-C-matrix-Inv-comparison-tp4655291p4655433.html

To unsubscribe from Octave/C++ matrix Inv() comparison, click here.
NAML

---

View this message in context: Re: Octave/C++ matrix Inv() comparison
Sent from the Octave - Maintainers mailing list archive at Nabble.com.