define objective function based on regression model

3 次查看（过去 30 天）
Khatere Dakhili
Khatere Dakhili 2022-11-28
回答： Amal Raj 2023-3-23
Hello,
I'm trying to define an optimization problem. I have a model from Gaussian Process regression, and I have 3 inputs. I tried to define my objective function in two ways but none of them is working. The first one is:
p = optimvar("p","LowerBound",0,"UpperBound",1);
mio = optimvar("mio","LowerBound",-1,"UpperBound",1);
sigma = optimvar("sigma","LowerBound",0.01,"UpperBound",0.1);
% Set initial starting point for the solver
initialPoint.p = 0;
initialPoint.mio = 0;
initialPoint.sigma = 0.01;
% Create problem
problem = optimproblem;
problem.Objective = (GPR_model.predictFcn(array2table([p,mio,sigma],"VariableNames",{'p','mio','sigma'})) + 0.01804)^2;
The second one is with another function
function objective = objectiveFcn(p,mio,sigma)
temp = array2table([p,mio,sigma],"VariableNames",{'p','mio','sigma'});
yfit = GPR_hyperparameter_SV1.predictFcn(temp);
objective = (yfit + 0.01804)^2;
end
problem.Objective = fcn2optimexpr(@objectiveFcn,p,mio,sigma);
Does someone have a hint about how I should solve this problem?
Thank you very much in advance

请先登录，再回答此问题。

回答（1 个）

Amal Raj
Amal Raj 2023-3-23
In both cases, you are adding a constant term of 0.01804 to the predicted value before squaring it. It is not clear why you are doing this, but if it is part of your optimization objective, then it is fine.
To solve the optimization problem, you can use any optimization solver that is compatible with MATLAB's optimization toolbox, such as fmincon or patternsearch. You can specify the lower and upper bounds for the three optimization variables, as you have already done, and use the initial point provided to start the optimization process.

类别

在 Help Center 和 File Exchange 中查找有关 Choose a Solver 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by