0

Background I'm trying to optimize a set of 7 parameters which are the core configuration of some external engine (specifically Solr parameters). I already have an optimization function which grades the results.

The idea was to use Nelder Mead because the derivative is unknown and there are probably few local optimums.

The problem is that the parameters must be real numbers and Nelder-Mead creates fractions.

My questions:

1) Is there a known way to overcome this problem?

2) Are there different optimizations I can use which address my problem?

Thanks ahead

Eran H.
  • 101
  • Why is it a problem if you input rationals into a variable that takes reals? – Eric Stucky Aug 31 '14 at 18:12
  • Nelder mead adjusts the numbers by small fractions, if I round the value it will always use the same numbers. It doesn't feel right using Nelder mead here (My problem is discrete not continuous) – Eran H. Aug 31 '14 at 18:34
  • Is the merit function fast to compute? If so, what about approximate the gradient by finite differences and use a gradient-based descent method? – AndreaCassioli Sep 01 '14 at 18:52

0 Answers0