Obtain partial or intermediate (output) results from genetic algorithm

Dear all,

I defined a very long function to be optimized by the genetic algorithm. How do I obtain the output of

(predict(Model{1,3},x)-min(Y1))/(Y1_target-min(Y1)) and predict(Model{1,3},x)

when a optimal solution is found?

I gather that you have to access the intermediate results of the genetic algorithm in some way, but I am not sure if that is the right approach and how one should accomplish this when it is the right approach.

%% Loading the Data
load([pwd,'\taguchi_sample_irregular.mat'])

%% Parameters
% Predictors
X = taguchi_multi_variate_samples;
Y = [Payload(:) Work(:) Overflow(:)];
% Response 1 (i.e. payload objective)
Y1 = Payload;
% Response 2 (i.e. work objective)
Y2 = Work;
% Response 3 (i.e. spillage objective)
Y3 = Overflow;

%% Surogate modeling of the data set
% SVM regression model with polynominal kernel functions
Model{1,3} = fitrsvm(X,Y1,'KernelFunction','polynomial', 'KernelScale','auto');
Model{2,3} = fitrsvm(X,Y2,'KernelFunction','polynomial', 'KernelScale','auto');
Model{3,3} = fitrsvm(X,Y3,'KernelFunction','polynomial', 'KernelScale','auto');

%% Optimization
% Target values for each response/objective
Y1_target = max(Y1);
Y2_target = max(Y2);
Y3_target = max(Y3);

fun = @(x) [-((((0.*(predict(Model{1,3},x) < min(Y1))) + ...
    ((predict(Model{1,3},x)-min(Y1))/(Y1_target-min(Y1))).*(min(Y1) < predict(Model{1,3},x) & predict(Model{1,3},x) < Y1_target) + ...
    (1.*(predict(Model{1,3},x) > Y1_target))) .* ...
    ((0.*(predict(Model{2,3},x) < min(Y2))) + ...
    ((predict(Model{2,3},x)-min(Y2))/(Y2_target-min(Y2))).*(min(Y2) < predict(Model{2,3},x) & predict(Model{2,3},x) < Y2_target) + ...
    (1.*(predict(Model{2,3},x) > Y2_target))) .* ...
    ((0.*(predict(Model{3,3},x) < min(Y3))) + ...
    ((predict(Model{3,3},x)-min(Y3))/(Y3_target-min(Y3))).*(min(Y3) < predict(Model{3,3},x) & predict(Model{3,3},x) < Y3_target) + ...
    (1.*(predict(Model{3,3},x) > Y3_target))))^(1/3))];

[x, fval, exitflag,output] = ga(fun,4,[],[],[],[],[0 0 0 0], [200 200 200 200]);

[Alan Weiss]

Sorry, I don't quite understand your question. After you get the result x what happens when you try to run

predict(Model{1,3},x)

This was one of your questions, and I am sure that you have tried this, but what happens?

Alan Weiss

MATLAB mathematical toolbox documentation

[Tessa Kol]

Thank you for your reply. Lucky for me I already solved the problem. I indeed try to run:

predict(Model{1,3},x)

I get the predicted output of first objective corresponding to the optimal set. In total I combined three objectives into one function. "x" (small x) is the optimal set that is predicted by the genetic algorithm. If you are interested in more information about what I implemented in matlab, here is a useful link: https://www.itl.nist.gov/div898/handbook/pri/section5/pri5522.htm

It is called the desirability approach and the function I defined corresponds the formula for the overall desirability.