Warmstart about daqp

[darnstrom]

First of all, thank you for catching the typo in the documentation!

As for the warm start, the current way of doing it is, as you mention, by using update_ldp() followed daqp_solve(). How much work from previous solves can be saved depend on which problem data is updated. If the Hessian H and/or the normals of the constraints A change, the internal LDL factorization will have to be redone from scratch. Such a refacatorization is, however, not necessary if only the linear term in the objective function f and/or the constraint term in the constraints b changes.

If your problem falls in to the former case of changing H and/or A you either have to call
activate_constraints() after update_ldp() if you want to warm start from the previous active set, or call deactivate_constraints() if you want to cold start (this is not necessary if only f and/or b change, i.e., update_ldp() followed by daqp_solve() is sufficient in that case).

I have used the solver mostly in cases where only f and/or b need to be updated. Hence, the warm-starting procedure when H and/or A change is a bit more crude, but is, nevertheless, possible by running update_ldp() followed by activate_constraints() and daqp_solve() as mentioned above.

[julianwanner]

Thanks for clearing that up. My problem unfortunaltely falls into the first category, i.e. H and A change. So the LDL factorization will have to be redone. Nevertheless, this solver outperforms the other solvers I tried.