docNumCost
Centre for Vision, Speech & Signal Processing
 

Optimsation Cost Function Class Structure

NumCostC is a base handle class for optimisation cost functions. It is derived from NumFuncC and can therefore also be used as a numerical function.

In order to enable a degree of constrained optimisation the interface to this class is relatively complicated. There is a distinction made below between the full parameter vector P and the parameters used for optimisation X. X is a subset of P. The elements of P that are not in X are kept constant at values specified SetConstP. The initial value of X will also be obtained from SetConstP. The optimiser should only be aware of the vector X and the the function being optimised should only be aware of the vector P. NumCostC provides the interface between these two.

Two other basic cost functions are provided which allow for a degree of cost function algebra. They are NumCostInvertC which inverts another cost function and NumCostCombineC which finds a weighted average of a list of cost functions.

Author: Robert Crida, Created: 15/6/1998, Generated by DocCreate: March 20, 2001