How to characterize objective function in order to choose optimization method?

92 Views Asked by At

What are some good quantitative metrics to describe an objective function that may help choose which optimization method will work best?

E.g. if there are a lot of local minima we know that we should use something like a particle swarm optimization rather than a gradient descent method. But how to quantity this so that the metic is normalized and applicable to various problems with varying dimensionality?

I'm particularly interested in non-derivative numerical methods, but if there are any general metrics, I'd love to hear them too.