Brendan O'Donoghue
Brendan O'Donoghue
There is an attempt to fix this here: https://github.com/bodono/scs-python/pull/121 Hoping we can land this soon.
This should be fixed now, see discussion here: https://github.com/bodono/scs-python/pull/121
PR here: https://github.com/cvxgrp/scs/pull/319/files
You can warm-start just some of the variables I think, SCS will set the others to zero. But SCS is expecting the variables corresponding to the conic reformulation that CVXPY...
Oh also the MKL pardiso direct solver is much faster than the indirect solver, I would recommend using the direct solver if you can.
I am very surprised that the MKL pardiso solver requires about the same time as the indirect solver! It might be worth playing with the MPI settings (more threads or...
Can you post the entire output trace (or the very end)? I would be interested in seeing the total times / iterations for each. Are you controlling the number of...
Yes it's doing a factorization, I didn't realize that was done single-threaded, seems like a waste! That probably explains the slowdown for MKL then. In principle we could write an...
Thanks for posting this. For some reason when I run your python code I actually get 'solved' correctly form SCS (I tuned on verbose outputs from SCS to see what...
Can you post the output of what you get when you run with `verbose=True`? It might be that something is compiled wrongly in your setup somehow.