Quantcast
Channel: Intel® Software - Intel® oneAPI Math Kernel Library & Intel® Math Kernel Library
Viewing all articles
Browse latest Browse all 3005

Result inconsistency in MKL svd

$
0
0

I'm using the following MKL product:

Major: 2017

Minor: 0

Update: 2

on an Intel(R) 64 architecture.

I get inconsistencies in the results from the function LAPACKE_dgesvd. When I input the function with slightly different input matrices I get different V* outputs. These are not the expected inconsistencies due to the non-uniqueness of svd. In my case the input matrix represents a scatter of coordinates in 2D. So the matrix has two columns. The output from svd for the matrix V* should be the rotation matrix to the principal directions. However, in some of the cases I get a sign flipped version of the principal direction angle. This is not a rotation of 90 or 180 deg. compared to the correct principal direction, but rather a simple flipping of sign (23deg. instead of -23deg., i.e.). 

I'm using 'O' and 'S' for the second and third arguments, respectively. 

 

 


Viewing all articles
Browse latest Browse all 3005

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>