Uses the Convex.jl Problem Depot to test various optimization solvers
Author ericphanson
5 Stars
Updated Last
10 Months Ago
Started In
February 2020


Build Status Dev

Uses the Convex.jl Problem Depot to test various optimization solvers. See the docs for the test results. Uses for printing the test results to HTML tables (and GitHub actions to run everything). This is meant to help solver developers test against a variety of problems and see where things can be improved.

Note that some of the problems are fairly strange (e.g. only one feasible point), which can be difficult for solvers to handle.

Pull requests to tweak the parameters of the solvers or the exclusions, or add more solvers, are welcome. The exclusion methodology I am using is as follows:

  • exclude problems which belong to a class that the solver choses not to solve (e.g. Tulip is a linear programming solver, so SDPs, SOCPs, exponential cone problems, and mixed-integer problems are all excluded)
  • or problems which take an excessive amount of time to solve

Regarding the parameters, I am trying to use a single choice of parameters for the entire set of problems (as opposed to choosing different parameters for different problems).

I added all open-source (and non-MATLAB-based) solvers from, besides the LP and MILP-only ones. I've also added Mosek by running it locally.