Parameter Selection

[kirkrudolph]

Play with parameter selection

Previously, I didn't understand the meaning of the parameter selections.

Interface

Solver Settings API

Paper

Reading this section of the OSPQ paper helped my understanding.

Opinion

Personally, I dislike having an adaptive value. It probably makes sense as a default, general-purpose setting when the problem is unknown. However, for known applications (like MPC), it sounds like I should be able to set the Adaptive rho to False.

It also sounds like it'd be less buggy:

Note that the convergence of the ADMM algorithm is hard to prove in general if the ρ updates happen at each iteration.