Optimisation when state evaluation is costly.

Asked Dec 04 '19 at 09:45

Active Dec 04 '19 at 09:45

Viewed 16 times

Looking for advice on constrained optimization problem. Currently I construct a Lagrangian and use various descent algorithms, line searches, all the usual tricks. No matter what I use, I am limited by a very costly state-evaluation step.

Is there an optimization algorithm that I can implement to avoid state evaluation unless required?

asked Dec 04 '19 at 09:45

AngusTheMan

Can you calculate gradients and/or hessian? – Vasily Mitch Dec 04 '19 at 10:00
I can, but only finite difference, so again it is costly. – AngusTheMan Dec 04 '19 at 10:06
If evaluating the objective function is expensive, then optimization will be expensive. I don't see how one can make it cheap. – Rodrigo de Azevedo Dec 04 '19 at 10:27
If you can't calculate gradients (finite differences doesnt count), then the best choice for you is simplex method – Vasily Mitch Dec 04 '19 at 10:38
Why is the evaluation expensive? – Rodrigo de Azevedo Dec 05 '19 at 00:05

Optimisation when state evaluation is costly.

0 Answers0