Information Technology Reference
In-Depth Information
can indeed be used as the basis for creating yet more complex systems that
require the swift handling of huge quantities of random numerical constants.
We will describe these systems in chapters 7, 8, 9, and 10, but for now let's
see how this complex system manages to fine-tune the random numerical
constants in order to find good solutions to the problems in hand.
5.4 Special Search Operators for Fine-tuning the RNCs
The efficient evolution of such complex entities composed of different do-
mains and different components requires a special set of genetic operators.
The operators of the basic gene expression algorithm (mutation, inversion,
transposition, and recombination) are easily transposed to the GEP-RNC
system, but obviously the boundaries of each domain must be maintained
and the different alphabets must be used within the confines of the corre-
sponding domain. Mutation was as expected extended to the Dc domain,
although I decided to keep it separated from the mutation operator that con-
trols the rate of mutation in the head/tail domains in order to allow not only
a better control but also a better understanding of this important operator.
The inversion operator was also transposed to the GEP-RNC algorithm, with
the primary inversion operator being restricted to the heads of genes and a
new Dc-specific inversion restricted obviously to the Dc-domain. Both IS
and RIS transposition were transposed to the GEP-RNC algorithm and their
action continues to be obviously restricted to the heads and tails of genes.
However, a special transposition operator was created that operates within
the Dc alone and also helps with the circulation of the RNCs in the popula-
tion. Furthermore, a special mutation operator - direct mutation of random
numerical constants - was also created in order to directly introduce genetic
variation in the sets of RNCs; this way a constant flux of new numerical
constants is maintained throughout the adaptive process.
The extension of all forms of recombination to the GEP-RNC algorithm is
straightforward as their actions never result in mixed domains or alphabets,
and they are also very effective not only at reshaping the expression trees but
also contribute to the fine-tuning of the numerical constants. Gene transposi-
tion was also extended to the GEP-RNC algorithm and, like recombination,
its implementation is straightforward as it never gives rise to mixed domains
or alphabets. However, for it to be productive, the set of RNCs attached to
the transposing gene must also go with it.
Search WWH ::




Custom Search