These functions were optimized by MuGA configured with 128
MI in the main population. Selection is made by tournaments with
size 3. The operator selects 256 individuals for the mating pool,
and in this way MI with copies are guaranteed for the following
operators. Recombination is made by one point crossover operator
with probability 0.6. Rescaling was applied to maintain a
maximum total of copies in the main population of twice the
number of MI. These aspects of MuGA were kept constant over
all the experiments.
Mutation was used in the simple flip-bit with probability 1/l,
where l is the length of chromosome, and in MWM which adds to
1/l the probability produced by the wave function, with final result
limited to the maximum of 1. The used roughness of the wave
function (Equation 1) is 1 and thinness is 3 (Figure 2).
Replacement was experimented in three forms, generational
where the best offspring replace all parents, tournament between
one parent and one offspring, and MDR. In this experiment we
use binary tournaments in the MDR operator.
For reference we compare the results with a version of MuGA
where the number of copies is not used (NoCopy) that is similar to
Simple Genetic Algorithm (SGA). The difference is that in this
version (NoCopy) of MuGA only unique (non-repeated)
chromosomes are present in the population. Mutation in this
version is flip-bit and replacement is by tournament as previously
described.