search

yig / PySPQR

Python wrapper for the sparse QR decomposition in SuiteSparseQR.
Creative Commons Zero v1.0 Universal
34 stars 27 forks source link

Add rz decomposition routine #10

Closed SmithB closed 4 years ago

SmithB commented 6 years ago

When solving large linear systems, it's nice to be able to use parts of a QR decomposition without forming the Q matrix. Suitesparse allows a Q-less decomposition, which forms R and Z such that z=Q^Td. The modifications I made to the code let PySPQR return Q and Z for sparse systems.

yig commented 6 years ago

I am trying to decide if I like the function name. I looked to MATLAB for inspiration, but they don't have a different name. They decide what to compute based on the output arguments.

Some alternatives:

SmithB commented 6 years ago

I'd be happy either way. Thanks!

On Thu, Aug 9, 2018 at 2:10 PM, Yotam Gingold notifications@github.com wrote:

I am trying to decide if I like the function name. I looked to MATLAB for inspiration, but they don't have a different name. They decide what to compute based on the output arguments.

Some alternatives:

-

zr(), since z is returned first. e is never mentioned for some reason.

a keyword argument to qr() to specify what should be returned (compute_Q = False)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yig/PySPQR/pull/10#issuecomment-411898311, or mute the thread https://github.com/notifications/unsubscribe-auth/ACLG6F8YNma55h5CNc4t59sZ-_6GIEQmks5uPKVIgaJpZM4V2WTz .

yig commented 6 years ago

Sorry about the delay. Can you clarify what R and z are? Is it the "[C,R,p] = qr(A,B,'vector')" version on this MATLAB page: https://www.mathworks.com/help/symbolic/qr.html