Use Bayesian Optimization on Existing Database

I have a model that takes in 10 inputs and gives an output. I have used this to generate a large database. Now I want to use Bayesian optimization to get the best combination of those 10 inputs but have no idea where to go from here.

Any help is appreciated.

Answer 1

[Alan Weiss]

Do you want to query your database or run your model? If you want to just query your database, input the database into MATLAB and lok for the point that gives minimal output using the min function..

I do not understand why you restrict yourself to using Bayesian optimization, nor why you think that Bayesian optimization is appropriate for querying a database. So it is possible that I am misunderstanding you. Maybe you want to find good points in the database and then refine them further using some sort of search, such as Bayesian optimization or patternsearch or something else. But I really don't know what you need.

Alan Weiss

MATLAB mathematical toolbox documentation

[Sweet Caroline]

Your second set of statements is correct. The model is fairly expensive so I am wanting to to find good points in the database and do a further search/refinement using bayesian optimization.

Everything I have seen on matlab only really works with 1 or 2 input variables and I have 10.

[Alan Weiss]

All Optimization Toolbox™ and Global Optimization Toolbox solvers take just one input variable, but that variable can have any number of components, so it is not a restriction. Your control variable x could represent x(1),...,x(10). See Writing Scalar Objective Functions.

For a time-consuming objective function, if you have a recent Global Optimization Toolbox, I'd try Surrogate Optimization. Give the points that you have already evaluated in an InitialPoints structure; see options or Algorithm Control.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation