Convex.jl/Mosek/SCS setting callback

Specific Domains Optimization (Mathematical)
1

Hello all,

I am new to Julia so I apologize if this is trivial.

I am calling Mosek and SCS through Convex.jl and want to track how quickly these solvers are converging (for plotting purposes). Is it possible to pass a callback to Mosek/SCS through Julia that saves some information about each iterate?

Here is a MWE if that helps!

using LinearAlgebra
using Convex, SCS, MosekTools
using MathOptInterface
const MOI = MathOptInterface

function test()
    # problem copied from https://jump.dev/Convex.jl/stable/quick_tutorial/
    m = 4;  n = 5
    A = randn(m, n); b = randn(m, 1)
    x = Variable(n)
    problem = minimize(sumsquares(A * x - b), [x >= 0])

    values = []
    norms = []
    times = []

    # ==== made up code, how do I do this?
    function callback(iterate)
        push!(values, value(iterate))
        push!(norms, norm(iterate))
        push!(times, time(iterate))
    end
    problem.add_callback!(callback)
    # ==== end of made up code

    solve!(problem, SCS.Optimizer; silent_solver=true)

    # solve!(problem, Mosek.Optimizer; silent_solver=true)

    println(values)
    println(norms)
    println(times)
end

test()

Best,
Alex

2

No, it is not possible.

3

The only way I know how to do this would be to capture the solver’s output (to stdout) and parse it to extract the values you want to plot.

4

Got it. Unfortunately, capturing solver output doesn’t work for me as I want to plot other metrics.

Thanks for looking anyways!