Question for statisticians. I have process that obtains a specific property of a point in a two-dimensional space, if I calculated it in the whole plane it would give values as in the attached picture (not necessarily Gaussian, it is only illustrative). The thing is, I need to find the point where that property is greatest without having to calculate all of them. Is there an algorithm or numerical method to reduce the number of tests over the space and quickly find the hot spot?
Instead of searching for the maximum, look for the minimum of the negative of your function. If you can compute derivatives, you can try the “method of steepest descent” (a.k.a. gradient search method) or the Gauss–Newton method or the Levenberg–Marquardt method. Otherwise you can try the simplex method (Nelder & Mead).
Pablo Mtz: correct, if you are dealing with a linear problem. This, however, is not clear from Pablo Constantino 's question.
- Badges
- Science topic