Information Technology Reference
In-Depth Information
01234560123456
Ab
O
bcbcOAAcaa
c
-[1,6] = 7
Ab
c
bcbcOAAcaa
b
-[2,1] = 8
As you can see, two point-mutations had to occur in order to create this
perfect solution: the first one changed the “O” at position 2 in gene 1 into
“c”, shortening the parental sub-ET in two nodes; and the second, although a
less drastic change in terms of structure, managed to change the expression
encoded in sub-ET
2
by changing the “c” at position 6 in gene 2 into “b”.
It is worth noticing that, in this run, some individuals contain genes en-
coding one-element sub-ETs (namely, chromosomes 3 and 9 of generation 1)
as they have a terminal at the start position of one of their genes. Obviously,
all these genes were created by point mutation, as only mutation was used to
create genetic modification in this problem. But we will see that, in gene
expression programming, not only mutation but also inversion are capable
of introducing a terminal at the root of sub-ETs (see section 3.3 for a detailed
analysis of the genetic operators used in gene expression programming).
3.2 Fitness Functions and the Selection Environment
Fitness functions and selection environments are the two faces of fitness and
are, therefore, intricately connected. When we speak of the fitness of an
individual, on the one hand, it is always relative to a particular environment
and, in the other, it is also relative to the measure (the fitness function) we
are using to evaluate them. Consequently, the success of a problem not only
depends on the way the fitness function is designed but also on the quality of
the selection environment.
3.2.1 The Selection Environment
We know already that the selection environment is the input to the evolution-
ary system and, as such, the usefulness of the output, that is, the models
designed by the system, will depend greatly on the quality of the environ-
ment we choose for them to blossom.
Thus, the first requirement is a set of fitness cases representative of the
problem at hand. And the second consists of a well-balanced set in order to
avoid the creation of models that can only solve a partial, and often mar-
ginal, aspect of the overall problem. Indeed, if these models are allowed to
