Information Technology Reference
In-Depth Information
where
d
0
corresponds to the independent variable
a
. It is worth pointing out
how unconventional this program is and how the algorithm got around the
absence of numerical constants by finding very creative forms of represent-
ing them. Notwithstanding, this model is a very good approximation to the
target function (5.15) as both the R-square on the testing set and the com-
parison of the plots for the target function and the model show (Figure 5.13).
25
20
Target
15
Model
10
5
0
-1.2
-0.6
0.0
0.6
1.2
-5
-10
Figure 5.13.
Comparison of the target function (5.15) with the model (5.19) evolved
by the GEA-B algorithm, that is, without explicitly using numerical constants. The
R-square was evaluated over a testing set of 100 random points
and is equal to 0.999977473.
The best solution created with the GEP-NC algorithm was found in gen-
eration 4220 of run 52. Its genome is shown below (the sub-ETs are linked
by addition):
0123456789012
*a++*5a5a1314
L*aa1+21141a2
SSSSS5a1a41a2
*a++aaa15544a
/EE+SEa2521a3
(5.20a)
It has a fitness of 999.635 and an R-square of 0.999993481 evaluated over
the training set of 20 fitness cases and an R-square of 0.999990385 evalu-
ated against the same testing set used in the GEA-B experiment, and there-
