Suppose we have a convex conic optimization problem (either SOCP or SDP) and its DCP model. Is it computationally advantageous to use CVX with the DCP model, or would we gain any computational benefits if we manage to derive the standard form of our problem and call a solver directly?
I don’t know how the solver time and efficiency would compare between using CVX and calling the solver directly - I suppose that could be problem dependent, and depends on how nice for the solver your formulation is vs. CVX’s. But calling the solver directly will save the CVX modeling time (incurred on every problem instance), at the expense of your human modeling time (perhaps incurred once per problem type).
If you want the fastest overall computer time, you are unlikely to get it from CVX, which is designed to save human modeling time, and perhaps reduce error propensity.
I appreciate you taking the time to respond. Thank you.