Information Technology Reference
In-Depth Information
HUMIDITY
high
normal
OUTLOOK
TEMPERATURE
sunny
overcast
rainy
hot
mild
cool
No
WINDY
Yes
Yes
Yes
Ye s
true
false
Yes
Yes
Figure 9.2.
A more colorful rendition of the decision tree encoded in gene (9.2).
chromosomes are expressed and their fitnesses evaluated against a particular
training set such as the data presented in Table 9.1. And according to fitness,
they are selected to reproduce with modification. The modification mechanisms
are exactly the same that apply in a conventional unigenic system, namely,
mutation, inversion, IS and RIS transposition, and one- and two-point re-
combination (see section 3.3, Reproduction with Modification).
So, as you can see, nothing but the usual tricks of evolution will be used to
design GEP decision trees. Let's now see how these decision trees learn by
studying a design experiment in its entirety.
9.1.2 A Simple Problem with Nominal Attributes
For the simple problem of this section we are going to use the play tennis data
presented in Table 9.1 and, therefore, the set of attributes will consist of A =
{O, T, H, W} and the set of terminals will consist of T = {a, b}. The fitness
function will consist of the number of hits, that is, the number of instances
correctly classified. As usual for this kind of illustrative problem, we are going
to use small populations of just 20 individuals so that we can analyze the
complete evolutionary history of a successful run. The complete list of pa-
rameters used in this experiment is given in Table 9.2.
