runcutest.jl

using CUTEst, JSOFilter, NLPModelsIpopt, Plots, SolverBenchmark, SolverTools
pyplot()

function runcutest()
  #ps = CUTEst.select(max_var=300, min_con=1, max_con=300, only_free_var=true, only_equ_con=true, objtype=2:6)
  #ps = CUTEst.select(max_var=300, min_con=1, max_con=300, only_free_var=true, objtype=2:6)
  ps = CUTEst.select(max_var=300, min_con=1, max_con=300, objtype=2:6)
  #ps = CUTEst.select(max_var=2, min_con=1, max_con=2, objtype=2:6)
  problems = (CUTEstModel(p) for p in ps)

  problems = setdiff(problems, ["GMNCASE1"])
  solvers = Dict(:filter => (nlp; kwargs...) -> filtersqp(nlp, max_time=60.0, max_eval=100_000, kwargs...),
                 :ipopt  => (nlp; kwargs...) -> ipopt(nlp, print_level=0, max_cpu_time=60.0, kwargs...))

  # Precompiling
  for (name, solver) in solvers
    nlp = CUTEstModel("HS7")
    solver(nlp)
    finalize(nlp)
  end

  stats = bmark_solvers(solvers, problems)
  for (name, df) in stats
    open("$name.md", "w") do io
      markdown_table(io, df, cols = [:name, :nvar, :ncon, :status, :objective, :elapsed_time, :neval_obj, :dual_feas, :primal_feas])
    end
    open("$name.tex", "w") do io
      latex_table(io, df, cols = [:name, :nvar, :ncon, :status, :objective, :elapsed_time, :neval_obj, :dual_feas, :primal_feas])
    end
  end
  df = join(stats, [:status, :neval_obj, :dual_feas, :primal_feas],
            invariant_cols = [:name],
            hdr_override = Dict(:neval_obj => "#f", :dual_feas => "dual", :primal_feas => "prim"))
  open("joined.md", "w") do io
    markdown_table(io, df)
  end
  open("joined.tex", "w") do io
    latex_table(io, df)
  end

  return stats
end

function profile(stats)
  costs = [df -> (df.status .!= :first_order) * Inf + df.elapsed_time;
           df -> (df.status .!= :first_order) * Inf + df.neval_obj]
  costnames = ["Elapsed time", "Number of objective evaluations"]
  profile_solvers(stats, costs, costnames, width=600, height=600)
  png("profiles")
end

stats = runcutest()
profile(stats)