OptimMPB.jl

This package provides glue code for using algorithms in Optim.jl using the MathProgBase interface. This is still a work in progress and ultimately the code here should belong to Optim.jl.

This simple example illustrate how things work:

using OptimMPB  ## OptimMPB reexport Optim and MathProgBase

type Rosenbrock <: MathProgBase.AbstractNLPEvaluator
end

MathProgBase.features_available(d::Rosenbrock) = [:Grad]
MathProgBase.eval_f(d::Rosenbrock, x) = (1-x[1])^2 + 100*(x[2]-x[1]^2)^2
function MathProgBase.eval_grad_f(d::Rosenbrock, gr, x)  
  gr[1] = -2*(1-x[1]) - 400*x[1]*(x[2]-x[1]^2)
  gr[2] = 200*(x[2]-x[1]^2)
end

s = OptimSolver(BFGS, show_trace = true, store_trace = true, extended_trace = true)
m = MathProgBase.NonlinearModel(s)
MathProgBase.loadproblem!(m, 2, [-1,-1], [2.,2.], :Min, Rosenbrock())
MathProgBase.setwarmstart!(m, [0.1,-0.1])
MathProgBase.optimize!(m)

MathpogBase.getobjval(m)
MathProgBase.getsolution(m)

If there some of the variables are bounded, the solver use Fminbox to deal with the bounds. If instead all the variables are unbounded, i.e. all(getvarLB(m))==-Inf and all(getvarUB(m))==+Inf then the regular solver is used.

If MathProgBase.features_available(d::AbstractNLPEvaluator) = [] then numerical derivatives are used to obtain gradient information. To use automatic differentiation pass autodiff=true to the OptimSolver constructor call.