Bayesian Optimization Results Evaluation

I am trying to learn and understand Bayesian Optimization. My code is working like in the documentation page but what is the difference between best observed feasible point and best estimated feasible point? Which result should I consider? Thanks for the help.

Answer 1

[Alan Weiss]

The difference is that the algorithm makes a model of the objective function, and this model assumes that observations can contain noise (errors). So the best observed feasible point is the one with the lowest returned value from objective function evaluations. The best estimated feasible point is the one that has the lowest estimated mean value according to the latest model of the objective function.

If your objective function is deterministic, then you can set the 'IsObjectiveDeterministic' name-value pair to true, and then these two points are likely to coincide.

Alan Weiss

MATLAB mathematical toolbox documentation

[MByk]

Thank you very much.

[muhamed ibrahim]

Does bayesian optimization in matlab consider observations are noisy by default? what can I do to run the bayesian optimizer in matlab cosidering the observations are noise free?

[Alan Weiss]

Yes. See the IsObjectiveDeterministic option.

Alan Weiss

MATLAB mathematical toolbox documentation

[muhamed ibrahim]

Thank You so much. I still need to have the bayesian optimization algorithm stop the iterations not after a specific number of iterations like (30 as default) but to stop when observed objective value reaches a specific value like 0.

[Alan Weiss]

To stop an optimization early, use the OutputFcn name-value pair. For details, see Bayesian Optimization Output Functions.

Alan Weiss

MATLAB mathematical toolbox documentation

[muhamed ibrahim]

regardin what you typed "and this model assumes that observations can contain noise (errors).

"How does Matlab compute this amount of noise? is it an arbitarary value?