Output of optimize() in Optim package

[Oliver Lylloff]

Hello all,

I'm trying to get acquainted with the Optim package - so far I think everything looks very interesting.

I would like to get the solution vector of each iteration as an output (e.g. for plotting) - how do I do that?

Trying the Rosenbrock example from https://github.com/JuliaOpt/Optim.jl with extended_trace option

optimize(f,g!,[0.0,0.0],method=:gradient_descent,extended_trace=true)

prints the solution vector x at each iteration but I can't seem to access it and store it. Any ideas?

Best, 

Oliver

[John Myles White]

Hi Oliver,

The result of optimize is an object with a field called minimum that has the solution.

Try something like the following:

julia> res = optimize(x -> (10.0 - x[1])^2, [0.0], method = :gradient_descent)
Results of Optimization Algorithm
* Algorithm: Gradient Descent
* Starting Point: 0

* Minimum: 10.000000000118629

* Value of Function at Minimum: 0.000000
* Iterations: 1
* Convergence: true
* |x - x'| < 1.0e-32: false
* |f(x) - f(x')| / |f(x)| < 1.0e-08: false
* |g(x)| < 1.0e-08: true
* Exceeded Maximum Number of Iterations: false
* Objective Function Calls: 4
* Gradient Call: 4

julia> res.minimum
1-element Array{Float64,1}:
10.0

— John

[Oliver Lylloff]

Hi John,

Thanks for the fast answer. 

I still don't think I get the output I want. The res.minimum only return the converged solution (one vector), however I would like to get the minimum for all 1 to N iterations (N vectors). Setting store_trace=true gets me closer, since res.trace prints the information I'm seeking but not as an field (res.trace.x).

Hope it makes sense.

Best,

Oliver

[John Myles White]

Hi Oliver,

If you’re looking for the current state at each iteration, you want to check res.trace.states.

One way to field out this kind of information is to use the names function:

julia> names(res)
15-element Array{Symbol,1}:
:method
:initial_x
:minimum
:f_minimum
:iterations
:iteration_converged
:x_converged
:xtol
:f_converged
:ftol
:gr_converged
:grtol
:trace
:f_calls
:g_calls

julia> names(res.trace)
1-element Array{Symbol,1}:
:states

Of course, you still need to figure out what the fields mean, but we’ve tried to use sensible names for the fields in Optim.

— John

[Oliver Lylloff]

Thanks John, 

Still trying to get a hold of the basics but this seemed to do the trick:

res = optimize(f,g!,[0.0, 0.0], method = :gradient_descent,store_trace=true,extended_trace=true);

n = res.iterations+1;

x_iter = zeros(2,n);

for i = 1:n

  x_iter[:,i] = res.trace.states[i].metadata["x"];

end

Best, 

Oliver