Object‐Oriented Programming of Optimizers – Examples in Scilab

Optimization algorithms are generally implemented following a functional paradigm that places the optimization method at the top of the hierarchy. Consider for instance the package Scilab [CAM 06], where an objective function f is optimized by a single call to a procedure that performs the whole optimization process, without any interference of the user:-> [yopt, xopt] = optim(f, x0) Remark – Objective functions and constraints are calculated from the outputs of a (programming) function or an external software, both of which we call simulators. To keep the text readable, the words " objective function " and " simulator " are used here as a generalization of objective functions and constraints. The ideas presented also apply to constraints and multiple objectives. Such a structure turns out to be very restrictive to implement more versatile strategies , such as changing the optimizer or f during the optimization, which may be needed, e.g., to couple local and global optimizers (e.g., [HAR 05]) or to replace high fidelity simulations by metamodels (see Chapter 5).