Replies: 2 comments 1 reply
-
|
@JAT38 This is a great question. You're right, there is no mathematical reason stopping kron to be DPP compliant.
I think 3. could also be required for the N-dimension support that @Transurgeon and @SteveDiamond are working on. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for your thoughts. This is not a showstopper for me. It's nice to see CVXPy getting better and better. |
Beta Was this translation helpful? Give feedback.
-
|
Just following up to say I came back to this problem a couple days ago. I should have realized before that it's all doable with vec, outer products, and reshape. All of these are DPP and not noticeably less efficient than kron. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hello,
I am solving a biconvex problem by alternating between two convex problems. I'd like to use parameters to speed up the solution of both.
One of the problems is not DPP because I take some Kronecker products (cvxpy.kron) between parameter matrices and constant numpy matrices. It seems to me the result of this should be parameter affine. But I see in the code that the Kronecker product is never DPP.
One workaround is to use for loops with a bunch of hstacks and vstacks, but this is very slow.
Any reason Kronecker products aren't DPP? I'm basically using them for what repeat does in Numpy. Any obvious tools I'm missing here?
Josh
Beta Was this translation helpful? Give feedback.
All reactions