Page 192 - Data Science Algorithms in a Week

176 Luis Rabelo, Edgar Gutierrez, Sayli Bhide et al.

Figure 1: Basic cycle of EAs.

The next step is reproduction where offspring are derived from the selected
individuals by applying the reproduction operations. There are usually three different
reproduction operations: 1.) mutation, which modifies with some probability the original
structure of a selected individual, 2.) reproduction (i.e., cloning of some individuals to
preserve features which contribute to higher fitness), and 3.) crossover, which combines
two chromosome instances in order to generate offspring. Blum et al. (2011) described
that “whether the whole population is replaced by the offspring or whether they are
integrated into the population as well as which individuals to recombine with each other
depends on the applied population handling strategy.”
The most popular EAs are Genetic Algorithms (GAs), Genetic Programming (GP),
Evolutionary Strategies (ES) and Evolutionary Programming (EP). The basic idea behind
GP is to allow a computer/machine to emulate what a software programmer does. The
software programmer develops a computer program based on objectives and gradual
upgrades. Langdon et al. (2010) stated that GP “does this by repeatedly combining pairs
of existing programs to produce new ones, and does so in a way as to ensure the new
programs are syntactically correct and executable. Progressive improvement is made by
testing each change and only keeping the better changes. Again this is similar to how
people program, however people exercise considerable skill and knowledge in choosing
where to change a program and how.” Unfortunately, GP does not have the knowledge
and intelligence to change and upgrade the computer programs. GP must rely on
gradients, trial and error, some level of syntactic knowledge, and chance.