Closed imciner2 closed 4 years ago
I believe that the behaviour here is partly luck. If I change the call to matlab to use a poor choice of initial
rho
and a tight tolerance, I can force to solver to use more iterations and adaptrho
. In that case the behaviour is not the same, since it appears to ignore some of my settings.Example : change the call to the
setup
function tosolver.setup(P, q, A, l, u, 'verbose', true, 'rho', 1e6, 'eps_rel', 1e-8, 'adaptive_rho_interval',50);
I think this is related to the use of the adaptive_rho_interval
as a whole rather than this example. I was just porting over the basic example to the matlab solver so it is easier to debug and the example problem matches the example given in the source code.
The problem with the adaptive_rho_interval
should probably be discussed in a new issue in the main repository.
Ping. I think this can be merged in, since it wasn't actually trying to solve the rho
behavior, it was just putting a simple example into the interface.
This creates a deterministic example that uses both code generation and the matlab solver and that matches the osqp_demo.c example given with the source code.