Infinite Objective Function Values in surrogateopt

How does the surrogateopt algorithm handle Inf values (I use it as return value for errors using try/catch) of the objective function? Will it distort the surrogate and negatively impact the optimization?

[Mario Malic]

Hi,

It's questionable, under algorithms, there's a link that leads to detailed description of procedure. The surrogateopt (may or will, I haven't dug deep into it) call fmincon which can't handle NaN or Inf values according to this answer from long time ago.

It's best to wait for advice from someone who knows this solver very well. Below is not well-informed suggestion that you could do.

Maybe you can set the value of objective function to be the maximum of previous ones or few times higher, or set the value of constraint violation for that evaluation to be higher than the constraint tolerance.