Bayesian Optimization Output Functions within Matlab App

Hi,

I'm trying to implement an optimization process, based on the bayesopt method, integrated into a GUI, made with application designer. Everything works fine except for the output function, which I'm having trouble with the arguments.

In the help section, the description is relatively straight forqward as

results = bayesopt(fun,vars,'OutputFcn',@outputfun,...

implying the automatic arguments results and state.

However, when calling a function within an app, I need to also supply the argument app. I tried several ways, including an anonymous function like

outFunc = @(x)outputFunction(app,x.results,x.state);

but just can't make it work. Depending on the formatting and whether or not uing the anonymous function, I always get one of the following two errors:

• Dot indexing is not supported for variables of this type.
• Too many input arguments.

Maybe (and even hopefully) it's an obvious mistake with an easy fix. Does anyone know a solution to this?

Thank you!

[Ulrich Streppel]

is there any solution in the meanwhile? I have the same problem. How can the app argument integrated into an output function in app designer?

Answer 1

[Alan Weiss]

I don't know what you mean "when calling a function within an app, I need to also supply the argument app."

However, maybe I can help. As explained in Bayesian Optimization Output Functions, an output function takes the signature

stop = outputfun(results,state)

So you might need the signature

@(results,state)outputfun(results,state)

But as far as I know that is the same as

@outputfun

so I am not sure that my advice is going to change anything for you.

Alan Weiss

MATLAB mathematical toolbox documentation

[Jon]

Thanks for your reply and sorry, if I didn't describe it clearly enough.

The function should be called, including a third (or rather first) argument "app", handing over the app handle/object, as in

outFunc = @(results,state)outputFunction(app, results, state);

because the function itself also should be initialized like the following

function outVal = outputFunction(app, results, state)

This is of course only true, if inside an app designer environment. Otherwise, to my understanding there is no way of accessing the private data, callbacks and functions of the GUI and app itself. Now how can I use the bayesian output function implementation without errors regarding input arguments?

Outside of an app, with just the result and parameters given, it's a simple imlementation and working perfectly fine for me.