[GitHub] [incubator-mxnet] zboldyga commented on issue #14360: supporting matrix inversion and determinant

[GitBox <gi...@apache.org>]

zboldyga commented on issue #14360: supporting matrix inversion and determinant 
URL: https://github.com/apache/incubator-mxnet/issues/14360#issuecomment-512293397
 
 
   @arcadiaphy I got distracted the past few months! I just picked this up today. Looked over the changes and will start coding tomorrow, should have a PR open within a few days.
   
   Immediately I realized that using the Potri and Potrf approach (which relies on LAPACK's Cholesky factorization) only works for symmetric positive-definite matrices. For matrices that meet these criteria, it's (probably) the fastest way to get the inverse. 
   
   That said, I'm guessing most cases where the inverse is needed will violate this requirement, so we'll need to create an inverse function based on the LU factorization, which LAPACK also supports (this should be easy). LU factorization is a good, relatively quick approach.
   
   Since the Cholesky operators were already added into the library, I'm going to assume those are needed somewhere. I'll leave those as is. 
   
   I'll approach the inverse, determinant, and logdet based on alternate LAPACK routines.
   
   As for MAGMA, I don't think this thread will be the appropriate place to introduce that. This may be a good candidate for discussion on the V1.6 release roadmap (here's V1.5: https://github.com/apache/incubator-mxnet/issues/14619, as of writing V1.5 is closed and V1.6 will probably open soon).

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services