passing variable through pattern search iterations

Hi everyone!

I'm using pattern search to solve a minmax problem. I know that pattern search:

1) Starts witha a polling phase where it polls the points in the current mesh by computing their objective function values,

2) it groups all the values of the objective functions and it select the mesh case with highest objective function value,

3) it moves the mesh in the last successful poll point (or it leaves the central mesh point as before) and starts again from 1),

4) this continues untill convergence is reached (possibly).

My question is: Is it possible to pass a variable from the best objective function (point 2) to the next polling phase (point 3)?

Many thanks!

[Ameer Hamza]

Can you clarify what the intended outcome is. Are you trying to read the intermediate values of optimization variables after each iteration of pattern search?

[Andrea Agosti]

Dear Ameer,

thanks for your answer. Yes you understood correctly, between each iteration of the pattern search I want to be able to read with the value of the objective function, also another variable. This variable will be later passed for the next iteration of pattern search.

Thanks for your help

[Venus liria silva mendes]

Hi everyone

%% Modify options setting

my example problem:

[combination, custototal, exitFlag, Output, population, scores] = ga (@ smc09v7AG_01, n_vars, A, b, Aeq, beq, LB, UB, NON_linear, Integral_variables, settings)

'' population '' I'm not sure if all individuals from all generations or just the last one return. And the "scores" returns the evaluations of each one.

Hope it works!

https://www.mathworks.com/help/gads/genetic-algorithm-options.html

Answer 1

[Ameer Hamza]

Following code shows how to get the information from each iteration of patternsearch

global x_iterations y_iterations
x_iterations = [];
y_iterations = [];

obj_fun = @(x) sum(x.^2.*exp(x.^2).*abs(log(x+1)));
opts = optimoptions('patternsearch', 'OutputFcn', @myOutFcn);
[x_final, f_final] = patternsearch(obj_fun, rand(1,10), [], [], [], [], [], [], [], opts);

function [stop, options, optchanged] = myOutFcn(optimvalues, options, flag)
    global x_iterations y_iterations
    x_iterations = [x_iterations; optimvalues.x];
    y_iterations = [y_iterations; optimvalues.fval];
    stop = false;
    optchanged = false;
end

This page show how to define the outputFcn to get more detail for each iteration of the optimization algorithm: https://www.mathworks.com/help/gads/pattern-search-options.html#f14623

[Andrea Agosti]

thanks, this worked perfectly

[Ameer Hamza]

Glad to be of help.

[Zakaria]

Does this methodology work with Genetic Algorithm optimizioation ?

I noticed that the structure of the OutputFcn is not the same.

[Ameer Hamza]

Yes, it is different. Please check my answer on your question.