Reviewed by
Ray Wyatt
School of Anthropology, Geography and Environmental Studies, University of Melbourne
A huge amount of work has gone into producing this book. This is why it stands as a very comprehensive coverage of optimisation techniques that have been applied to agriculture. It will therefore prove useful as a reference guide but it is difficult to see why else anyone would want to purchase it.
For instance, one would not buy it in order to learn more about how the different optimisation techniques actually work. The author assumes that the reader is already knowledgeable about such techniques, and while some useful insights are given about subtleties of different methods, a slim volume like this one is simply too short of space to play the educative role properly.
Nor would one purchase this book to learn which technique works best. Many contradictory results are presented, with instances given where genetic algorithms outperform hill climbing; and vice versa, along with instances where simulated annealing outperforms evolutionary programming, and vice versa, or simplex is best/worst, and so on. Thus, no definite conclusion is reached, and it is further pointed out how various authors frequently find that their favoured technique is the best one, via imprecise testing within problem domains that will favour their approach anyway.
For example, certain techniques (such as hill climbing) are more likely to perform well when the utility surface of the solution space is smooth. By contrast, the genetic algorithm (or even simplex optimisation) can be more effective when applied to those problems whose domain of feasible solutions is marked by violent changes in the utility surface.
Finally, one would hardly want to buy this book as a guide to applications within one particular field. The term "agricultural systems" is simply too wide. It addresses problems such as optimisation of crop types across a landscape, the minimisation of nutrient runoff from paddocks, maximising profit through scheduling of dairy activities and scores more. As such, the field of application considered in the book is not a field at all; it is several fields, and each one will tend to see different techniques work best for the applications it is interested in.
In other words, the field of application considered is so polyglot that no conclusions can be safely drawn about what the best technique is for the discipline. It depends on what discipline (and problem type) is being considered. Eventually therefore, the author makes his own recommendation as to the "best" method – a form of genetic algorithm. This recommendation turns out to be based on many unstated assumptions, thereby committing the same sin as the author accuses others of committing whenever they write up the virtues of their own (favoured) approach.
We are left with something that looks like the literature review from a thesis‚ a broad-ranging survey of work done in far too wide a field, designed to impress the examiner with its breadth of knowledge and meant to "choose the territory" for what will be achieved in the core of the thesis to follow. But just as universities do not give a degree for a literature review, people are seldom likely to buy any book that is mainly a literature review. It would be good to know what the author actually achieved, subsequent to this book, using his favoured algorithm, but the reader is never told.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2005