SOFTWARE REVIEW
This software review was published in the Journal of chemometrics
Simplex optimization has long been a popular and important area of chemometrics. Practitioners of the simplex optimization methodology have generally had to rely on home built software, though, because there has been no commercial software for doing simplex optimization, despite hundreds of research papers detailing its use and the regular offering of a short course dedicated to the topic of optimization. The MultiSimplex® package changes that unhappy situation by offering easy access to simplex optimization without the need to program.
MultiSimplex® is a sophisticated implementation of the simplex optimization method. Through pull-down menus the user has full control over the simplex optimization process. The parameters of the simplex search can be set through menus and the basic simplex or a modified simplex search can be set. Adjustable coefficients for simplex contraction, expansion and reflection and automated checking with a re-evaluation rule make the optimization very flexible and well suited to dealing with real data containing noise.
The most interesting feature of MultiSimplex® is its ability to optimize several response variables at the same time. Usually, simplex optimization is developed for optimization of a single response as a function of several control variables, but, by using fuzzy combination of the objectives, a multi-response optimization is made possible. MultiSimplex® has a set of three predefined membership functions that permit the user to select the relative significance of the responses being optimized, as well as how the significance of each response being optimized changes as the optimization progresses.
Other membership functions may also be used by changing a constant in the fuzzy optimization expression. The approach used in the MultiSimplex® software is novel; only one paper on the fuzzy combination of objectives in a simplex optimization has appeared previously. This approach seems to work well, at least on the small optimization problems that I used to test the software. The simplex optimization must be done by manual entry of responses and there is no capability of optimizing a user-defined function such as might be done for optimization of parameters in a non-linear function.
-However, both these functions are available in the more advanced OMM version, of the software-.
MultiSimplex® comes with two built in tutorial applications. One concerns a “simple” optimization with two response variables – where one is to be minimized and the other maximized – and two control variables. The second involves a more complex optimization with five control variables and three responses, one of which is to be maximized, one to be minimized and the third to be held to target value.
The software package generates the response values from an input set of control variables. The mechanics of these tutorial examples are discussed thoroughly, but there is relatively little in the way of background to explain how variables for the optimization were selected or decisions were reached on the membership functions suggested in the example.
The manual begins by covering the theory of the simplex optimization and theory of the fuzzy combination of the response variables. Then the commands and menus are presented. The two tutorials mentioned above are used to illustrate the workings of the software in practice. Chapters on the mechanics of defining a project and the optimization sequence follow. Chapter on the graphics and database – the storage of the sequence of control values and the responses – follow, then the manual concludes with brief coverage of user defined simplices and a brief section addressing more practical issues that may occur to the new user.
The MultiSimplex® package is an attractive package for users with a need to optimize one response or multiple responses. This well implemented version has a short learning curve.
