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
Den lørdag den 4. januar 2014 17.19.56 UTC+1 skrev 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
>
> On Jan 4, 2014, at 11:14 AM, Oliver Lylloff
> <[email protected]<javascript:>>
> wrote:
>
> > 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
> >
> > Den lørdag den 4. januar 2014 16.56.03 UTC+1 skrev 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
> >
> > On Jan 4, 2014, at 10:52 AM, Oliver Lylloff <[email protected]>
> wrote:
> >
> > > 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
> > >
> > >
> >
>
>
