How to Setup Surrogate Optimization with Data input

I have a function "SurrogateFunction" I would like to use Surrogate to optimse. However, i dont know how to setup the options.

Data1 and Data2 are double vectors (500x2 and 500x2 in size). They do not need to be optimize and are the data for testing.

Variable1, 2 are integer

Variable3 is a double

I would like to optimise variable 1, 2 and 3.

Variables have lower bound level of [60, 10 and 0.5] respectively

Variables have upper bound level of [900, 240 and 3.5] respectively

Using the optimse solver, i got the following code

% Set nondefault solver options
options = optimoptions('surrogateopt','Display','iter','PlotFcn',[]);
% Solve
[solution,objectiveValue] = surrogateopt(fun,lb,ub,intCon,[],[],[],[],options);
% Clear variables
clearvars options

function Value = SurrogateFunction(Data1, Data2, Variable1, Variable2, Variable3)
    %some code
end

This is the error i am getting:

Error using globaloptim.bmo.verifybounds (line 72)

INTCON must be in the range [1 3].

Error in globaloptim.bmo.createSolver

Error in globaloptim.bmo.BlackboxModelOptimizer (line 99)

self = globaloptim.bmo.createSolver(self,expensive,lb,ub, ...

Error in surrogateopt>createController (line 371)

controller = globaloptim.bmo.BlackboxModelOptimizer(expensive,lb,ub,intcon, ...

Error in surrogateopt (line 334)

controller = createController(expensivefcn,lb,ub,intcon,Aineq,bineq,Aeq,beq,options);

How do I config the optimization so it runs?

[Alan Weiss]

Please report the version of MATLAB you use.

Please report the values of fun, lb, ub, and intCon for your problem. (I suspect that your fun argument might be incorrect, meaning does not satisfy the requirements of an objective function.)

Please do not clear the options variable after creating it.

Alan Weiss

MATLAB mathematical toolbox documentation

[Clayton Leung]

Thanks for replying, to answer your question, here are the variable declaration. I know the @fun arguement is incorrect, but I dont know how to write it properly.

I am using the version R2021a.

NumerOfVariable = 3;
lb = [60,10,1];
ub = [900,240,4];
intCon = 1:900;
fun = @CC_Pairs_Surrogate;
options = optimoptions('surrogateopt','Display','iter','PlotFcn',...
            {'surrogateoptplot','optimplotfvalconstr','optimplotfval',...
            'optimplotconstrviolation','optimplotx'});
[solution,objectiveValue,exitflag,output,trials] = surrogateopt(fun,lb,ub,intCon,[],[],[],[],options);

Error Message:

Error using globaloptim.bmo.verifybounds (line 72)

INTCON must be in the range [1 3].

Error in globaloptim.bmo.createSolver

Error in globaloptim.bmo.BlackboxModelOptimizer (line 99)

self = globaloptim.bmo.createSolver(self,expensive,lb,ub, ...

Error in surrogateopt>createController (line 371)

controller = globaloptim.bmo.BlackboxModelOptimizer(expensive,lb,ub,intcon, ...

Error in surrogateopt (line 334)

controller = createController(expensivefcn,lb,ub,intcon,Aineq,bineq,Aeq,beq,options);

Thanks in advance.

Answer 1

[Alan Weiss]

The error is clear: you have just three variables, so the indices of the integer variables have to be in the range 1 through 3. I don't know what you are trying to do by specifying intCon = 1:900; but that specification makes no sense for the syntax.

In addition, I suspect that your definition of your objective function is incorrect. If you want help debugging your objective function, please provide the first line of the definition of CC_Pairs_Surrogate, which will be something like

function y = CC_Pairs_Surrogate(X,L,R)

Alan Weiss

MATLAB mathematical toolbox documentation

[Clayton Leung]

Hi @Alan Weiss,

I dont know how to declare my Data1 and Data2 variable, as they are vectors, not scalers.

I saw the example regarding using integers as variables, and they have a variable intCon, I just copy that. I don't know how to use this syntax properly, as my upper bound value goes to 900, I set my intCon to 1:900.

My Surrogate function is:

function Value = SurrogateFunction(Data1, Data2, Variable1, Variable2, Variable3)
    %some code
end

As I said, I have 2 Data Variables that do not change and do not need to be optimize, and I have 3 variables that i like to optimize.

Variable1 and Variable2 are integer variables

Variable3 is a double variable.

Variables have lower bound level of [60, 10 and 0.5] respectively

Variables have upper bound level of [900, 240 and 3.5] respectively

Thank you for your help

[Alan Weiss]

You have three variables. If all three should be integers, then set intCon = 1:3. If just variables 1 and 2 should be integer, set intCon = [1 2]. If you want the variables bounded as you say, set

lb = [60, 10, 0.5];
ub = [900, 240, 3.5];

As for your objective function, you need to write it this way:

function Value = SurrogateFunction(Data1, Data2, Variables)
Variable1 = Variables(1);
Variable2 = Variables(2);
Variable3 = Variables(3);
% Now the rest of your code
end

And when you specify fun in your surrogateopt call, do it like this. Firt get the Data1 and Data2 arrays into your workspace. Then execute

fun = @(Variables)SurrogateFunction(Data1, Data2, Variables);

For details, see Writing Scalar Objective Functions and Passing Extra Parameters.

Alan Weiss

MATLAB mathematical toolbox documentation

[Clayton Leung]

Thank you, I will try it as you said.