Hill climbing optimization

Constrained hill climbing optimization

Description

It is a hill climbing optimization algorithm for finding the minimum of a fitness function
in the real space. The space should be constrained and defined properly. It attempts steps on every dimension and proceeds searching to the dimension and the direction that gives the lowest value of the fitness function. It stops if all the neighboring function evaluations give a larger result than the lowest estimated one.

The user is able to define the precision (i.e. the step) that the algorithm will use. This makes it more flexible than the fminsearch optimization function that Matlab provides.

This algorithm is able to pass to the fitness function additional arguments if needed.

For examples type ?help hilloptions? and ?help hill?

• MATLAB release: MATLAB 7 (R14)
• Other Files: hilloptions.m, ; hill.m

[Karel]

Be warned; this script does not properly respect the bounds. In certain cases, it will select a point just outside of the bounds, due to a bug in the code (bounds are only taken into account when the fitness function is being evaluated, not when adding a step to the current best point).