I need to create a 3d plot with each step of fminsearch

Hi, here is my program

par0 = [0.005276, 0.5017, 88.6357];% initial values

fun_objetivo = @(par,pfrac)FunObjetivo(par);

options = optimset('MaxFunEvals',3000,'Display','iter','PlotFcns',@optimplotfval);

%We use here the Nelder-Mead method
par_optimos = fminsearch(fun_objetivo, par0, options);

I need help with some comands to create a 3d plot, what I want is a 3d plot with "par_optimos(1)" as X, "par_optimos(2)" as Y and "min f(x)" as Z , so I can see how two of the values of my vector "par_optimos" are optimized in each iteration.

Answer 1

[Walter Roberson]

options = optimset('MaxFunEvals',3000,'Display','iter','PlotFcns',@plot_12z);

and

function stop = plot_12z(X,OPTIMVALUES,STATE)
  persistent h
  
  stop = false;
  switch STATE
      case 'init'
        if isempty(h) || ~isvalid(h)
            fig = figure();
            ax = axes(fig);
            h = animatedline();
        end
      case 'done'
          if ~isempty(h) && isvalid(h)
            delete(ancestor(h), 'figure');
          end
      case 'iter'
          if isvalid(h)
             addpoints(h, X(1), X(2), OPTIMVALUES.fval);
          end
      case 'interupt'
          %no new data
  end
end

[Jorge Armando Vazquez]

I tried it, but for some reason, it appears error:

%Error using matlab.graphics.axis.Axes/set
%The UIContextMenu's parent figure is not the same as the
%parent figure of this object

%Error in callAllOptimPlotFcns (line 144)
        %set(handle,'UIContextMenu', cmenu);

%Error in fminsearch>callOutputAndPlotFcns (line 485)
            %stop =
            %callAllOptimPlotFcns(plotfcns,xOutputfcn,optimValues,state,varargin{:})
            %|| stop;

%Error in fminsearch (line 208)
    %[xOutputfcn, optimValues, stop] =
    %callOutputAndPlotFcns(outputfcn,plotfcns,v(:,1),xOutputfcn,'init',itercount,
    %...

%Error in Main_optim (line 7)
%par_optimos = fminsearch(fun_objetivo, par0, options);
 

Can you help me please . This is something really new for me.

[Walter Roberson]

Interesting, it looks like it creates a figure itself. So,

function stop = plot_12z(X,OPTIMVALUES,STATE)
  persistent h
  
  stop = false;
  switch STATE
      case 'init'
        if isempty(h) || ~isvalid(h)
            %calling function will have made a figure the current figure
            %and will have called subplot
            ax = gca;
            h = animatedline(ax);
        end
      case 'done'
          if ~isempty(h) && isvalid(h)
            delete(h);
          end
      case 'iter'
          if isvalid(h)
             addpoints(h, X(1), X(2), OPTIMVALUES.fval);
          end
      case 'interupt'
          %no new data
  end
end

[Jorge Armando Vazquez]

the error disappeared, but it keeps showing me a plot in 2d and I need one in 3d.

[Walter Roberson]

Try adding

view(3)

after the assignment to h in the 'init' section.

[Jorge Armando Vazquez]

Do you know how to make 3d plot with mesh plot function, with the same axis. I'd appreciate it.

[Walter Roberson]

What would be mesh'd ? You have a series of par_optimos(1) and par_optimos(2) values, and corresponding function values, but they would be scattered.

When there are at least 3 points, you could potentially do a triangulation and mesh plot of that, such as is shown at https://www.mathworks.com/help/matlab/ref/delaunaytriangulation.html

[Jorge Armando Vazquez]

what I need is a 3d plot but with a format similar of the meshgrid:

https://www.mathworks.com/help/matlab/ref/meshgrid.html

because I want to graph all the values that were between p0 and par_optimos, where "x" will be the values to get to par_optimos (1), "y" for the values of par_optimos (2) and z the error that occurred in each iteration, showing it with the meshgrid format. yeah I know sounds complicated but that's the way I was asked

[Walter Roberson]

?? For every minimization step, you are expected to graph for "all" of the values between p0 and par_optimos(1:2) ??

Or are you instead expected to run a minimization step without graphing in order to determine the location of the minima, and then graph for "all" of the values between p0 and the location of the minima ?

Or are you instead just expected to plot over a fixed range of x and y and highlight the minima observed over that range ?

[Jorge Armando Vazquez]

I think the second option sounds more reasonable

[Walter Roberson]

For certainty, please indicate whether "second option" is "Or are you instead expected to run a minimization step without graphing" or "just expected to plot over a fixed range of x and y" ? The last of those is the easiest. Especially if the objective function is vectorized.

[Jorge Armando Vazquez]

run a minimization step without graphing in order to determine the location of the minima, and then graph for "all" of the values between p0 and the location of the minima

[Walter Roberson]

Okay, in that case do not set a PlotFcn in the options.

Then, after the

par_optimos = fminsearch(fun_objetivo, par0, options);

have something like

N = 100;
xvec = linspace(par0(1), par_optimos(1), N);
yvec = linspace(par0(2), par_optimos(2), N);
p3 = par_optimos(3);
[X, Y] = meshgrid(xvec, yvec);
Z = arrayfun(@(x,y) fun_objectivo([x, y, p3]), X, Y);

surf(X, Y, Z, 'edgecolor', 'none');
xlabel('para_optimos(1)')
ylabel('para_optimos(2)');
zlabel('objective');

Note that this does not optimize at each point: it uses the same third optimization parameter as at the optimum it found.

[Jorge Armando Vazquez]

Look very good, just that this error appears:

Undefined function or variable 'fun_objectivo'.

Error in Main_optim>@(x,y)fun_objectivo([x,y,p3])

Error in Main_optim (line 15)

Z = arrayfun(@(x,y) fun_objectivo([x, y, p3]), X, Y);

[Walter Roberson]

Z = arrayfun(@(x,y) fun_objetivo([x, y, p3]), X, Y);

[Jorge Armando Vazquez]

I have been trying to run the program many times, and there appears no error but the program doesn´t stop running so I dont get the final plot, you think maybe if is by the way where for every minimization step, graph for "all" of the values between p0 and par_optimos(1:2) may works.

[Walter Roberson]

I do not see your code for FunObjetivo ?