Information Technology Reference
In-Depth Information
In the deterministic selection scheme, individuals are selected proportion-
ately to their fitnesses, but less fit individuals are excluded from the life
banquet. One can choose to exclude more or less individuals from the ban-
quet, and I have tested several deterministic selection schemes with different
exclusion levels. An exclusion factor
E
of 1.5 seems to be, most of the times,
the best compromise.
To apply this exclusion factor, the viability (i.e., the fitness divided by
average fitness) of each individual is multiplied by
E
. Imagine a round table
where the slices begin to be occupied by rank and proportionally to viability,
starting with the best individual and finishing when no more places are avail-
able on the table. The higher the exclusion factor the more individuals die
without leaving offspring and only the cream of the population reproduces.
Obviously, one must be careful with the degree of exclusion, otherwise the
genetic diversity might be excessively reduced and evolution might be seri-
ously compromised.
The problems chosen for this study are two test problems of sequence
induction. The first is the sequence (5.14) of section 5.6.1 and the second is
the more difficult sequence (12.1). The general settings chosen for both ex-
periments are summarized in Table 12.7.
Figures 12.15 and 12.16 compare the variation of success rate obtained
for the three selection schemes over a wide time span. In both experiments,
the tournament selection scheme is slightly inferior to both the roulette-wheel
and deterministic selection. Deciding between the roulette-wheel and
100
80
60
Roulette
Tournament
Deterministic
40
20
0
0
25
50
75
100
125
150
Number of generations
Figure 12.15.
Comparison of roulette-wheel, tournament, and deterministic
selection schemes on the simpler test sequence (5.14).
