How can I stop fminsearch in one iteration after a specific time?

I used fminsearch for fitting experiment data to a model. It has 5 parameres. For beter optimizing, I changed two parameters in every run of fminsearch but unfortunatly sometimes for some of parameters value, it stops in one iteration and doesn't go to another one or finish. I need to set a certain time for stopping these iterations and sending program to another one. How can I do this?

Thanks,

Answer 1

[Walter Roberson]

fminsearch will stop immediately if funvalcheck is set and the first iteration is nan or complex. Or if there is an error evaluating the function.

[fatemeh jafariani]

Thanks. I reduced the amount of 'MaxFunEvals' from 1e10 to 1e4 so now it runs quickly but I reached to NaN. how can I do to balance between both of the time of running and right answer? I really confused, I want to get out of this step of iteration that took long times :(

[Walter Roberson]

Did you check that your function returns non-nan on the initial conditions?

If your initial conditions return a finite value but at some point in searching you encounter nan, then the implication is either that you need constraints or else that you have reached a point where double precision is not sufficient to compute your values (for example, factorial(171)/factorial(170) should obviously be 171 but factorial(171) overflows double precision so you get infinity instead.)

fminsearch is not designed to have bounds of any kind. However, if you look in the File Exchange, John D'Errico has provided fminsearchbnd which supports upper and lower bounds.

[fatemeh jafariani]

Did you check that your function returns non-nan on the initial conditions? yes, I checked. you' re right, it returns a finite value but at some point in searching I encounter nan.

Thank you for your help, I will have a look in the File Exchange that you said.

[Walter Roberson]

dbstop if naninf

can be very useful.

However in practice it can turn out to be more useful to put a conditional breakpoint at the end of your function that tests whether the value you are about to return contains non-finite numbers. The conditional breakpoint approach avoids accidental triggers on the internals of the optimization functions.

Answer 2

[Thiago Henrique Gomes Lobato]

If when you say stop you mean it simply freezes and you don't get an error nor the iteration goes on than you probably have some while loop in your function that, for those parameters, simpy never ends or maybe a specific process that takes unreasonable many time for centain parameter combinations. Different from fminunc, for example, fminseach doesn't performs any line search, so as long as your function has constant time each iteration should take roughly the same amount of time. If it simply "freezes" then it is something in your function.

[fatemeh jafariani]

thank you for your help.