Information Technology Reference
In-Depth Information
compares with the simpler approach in which numerical constants are created
from scratch by the algorithm itself.
In order to do so, we are going to conduct two experiments with 100 runs
each and compare the performance of the two approaches by evaluating the
success rate on the simple problem of section 3.4. We are going to use again
the fitness function based on the selection range and precision (using an
absolute error of 100 for the selection range and 0.01 for the precision) as
this guarantees that all the solutions with maximum fitness match perfectly
the target function (3.19). The performance of both experiments and the pa-
rameters used per run are summarized in Table 5.1. And as you can see,
when the algorithm itself is used to create small integer constants from scratch,
the performance is considerably higher than in the case where the constants
Table 5.1
Performance and settings used in the simple function problem without using
numerical constants (
GEP
) and with a fixed set of NCs (
GEP-NC
).
GEP
GEP-NC
Number of runs
100
100
Number of generations
50
50
Population size
30
30
Number of fitness cases
10 (Table 3.2)
10 (Table 3.2)
Function set
+ - * /
+ - * /
Terminal set
a
a 0 1 2 3
Head length
7
7
Gene length
15
15
Number of genes
3
3
Linking function
+
+
Chromosome length
45
45
Mutation rate
0.044
0.044
Inversion rate
0.1
0.1
IS transposition rate
0.1
0.1
RIS transposition rate
0.1
0.1
One-point recombination rate
0.3
0.3
Two-point recombination rate
0.3
0.3
Gene recombination rate
0.3
0.3
Gene transposition rate
0.1
0.1
Fitness function
Equation (3.3a)
Equation (3.3a)
Selection range
100
100
Precision
0.01
0.01
Success rate
68%
10%
