Page 53 - 49A Field Guide to Genetic Programming

Chapter 5

Alternative Initialisations

and Operators in

Tree-based GP

The genetic programming system described in the preceding chapters is just
the beginning; in many ways it is the simplest thing that could possibly
work. Most of the techniques described in Part I date back to the late
1980’s and early 1990’s, a wide array of alternatives and extensions have
been explored since. A full catalogue of these would be far beyond the
scope of this book. The chapters in Part II survey a number of the more
prominent or historically important extensions to GP, particularly (but not
exclusively) in relation to the tree-based representation for programs.
We start, in this chapter, by reviewing a variety of initialisation strategies
(Section 5.1) and genetic operators (Sections 5.2 and 5.3) for tree-based GP
not covered in Part I. We also briefly look at some hybridisations of GP
with other techniques (Section 5.4).

5.1 Constructing the Initial Population

Koza’s ramped half-and-half method is the most common way of creating the
initial GP population (cf. Section 2.2, page 11). However, there are several
other ways of constructing a collection of random trees. In Section 5.1.2
we will briefly consider an unexpected impact of population initialisation.
There has also been some work with non-random or informed starting points
(cf. Section 5.1.3).

39