I saved a convex problem, not solved yet, created by using Convex.jl, in a MOF/JSON file, with the write_to_file function.
Now, I want to read that problem and start the opimization using the function Convex.solve!. However, there is no read_from_file implemented in Convex.jl. I know that I can read it and then solve it using JuMP, but I’d like to exploit warmstart, since I have a good initial solution. And I have not found an easy way to do it in JuMP, even though I am not an expert and could have missed something.
Is there a way to read the problem from file and exploit the warmstart?
I don’t know how easy it would be to have read_from_file in Convex.jl.
Do you have a warmstart for the full model that is in the .mof.json file? Or just the original user variables in Convex?
I checked and, actually, I do not have the warmstart in the .mof.json (i.e., there is no primal_start field). I do not know if this is an intended behavior of Convex.write_to_file or there is a problem somewhere.
Just to give a little bit of context, my problem is a MISOCP. I construct the model using Convex.jl, then I set the value of the solely integer variables with set_value!, and then I use Mosek to solve the problem with the attribute MIO_CONSTRUCT_SOL so that it can construct a feasible solution from the partial integer solution, reducing so the runtimes.
Why are you saving to a file? Why not just build and solve from Convex.jl?
Actually, I am working with the Mosek support to reproduce a bug (causing the solver to freeze and BSoD the computer), and they’d like to have a model file. Plus, since it is working material, I do not want to provide my full software.