Information Technology Reference
In-Depth Information
In this classification task, for gene expression to be complete, the output of
each sub-ET must first be converted into 0 or 1 using a rounding threshold,
below which the output is converted into 0, 1 otherwise. Then the sub-ETs
must be subjected to the following rules in order to determine the class:
IF (Sub-ET
1
= 1 AND Sub-ET
2
= 0 AND Sub-ET
3
= 0), THEN Class 1;
IF (Sub-ET
1
= 0 AND Sub-ET
2
= 1 AND Sub-ET
3
= 0), THEN Class 2;
IF (Sub-ET
1
= 0 AND Sub-ET
2
= 0 AND Sub-ET
3
= 1), THEN Class 3.
Let's make this more concrete with a simple example, a small sub-set of
the iris dataset (Fisher 1936) where the first five samples of each type of iris
in the original dataset are used (Table 2.1). Indeed, the program (2.12) above
was created using the training samples shown in Table 2.1. And, as you can
easily confirm with the help of Figure 2.4, this model classifies all the sam-
ples correctly using a rounding threshold of 0.5.
So, for problems with multiple outputs, multiple sub-programs are en-
coded in the chromosome and the “organism” is the result of an intricate
collaboration between all sub-programs, in which each sub-program is en-
gaged in the discovery of a particular facet of the global problem.
Table 2.1.
The iris sub-set.
Sepal length (a)
Sepal width (b)
Petal length (c)
Petal width (d)
Type
5.1
3.5
1.4
0.2
class 1 (seto.)
4.9
3.0
1.4
0.2
class 1 (seto.)
4.7
3.2
1.3
0.2
class 1 (seto.)
4.6
3.1
1.5
0.2
class 1 (seto.)
5.0
3.6
1.4
0.2
class 1 (seto.)
7.0
3.2
4.7
1.4
class 2 (vers.)
6.4
3.2
4.5
1.5
class 2 (vers.)
6.9
3.1
4.9
1.5
class 2 (vers.)
5.5
2.3
4.0
1.3
class 2 (vers.)
6.5
2.8
4.6
1.5
class 2 (vers.)
6.3
3.3
6.0
2.5
class 3 (virg.)
5.8
2.7
5.1
1.9
class 3 (virg.)
7.1
3.0
5.9
2.1
class 3 (virg.)
6.3
2.9
5.6
1.8
class 3 (virg.)
6.5
3.0
5.8
2.2
class 3 (virg.)
