Evolution Strategy

    An Evolution Strategy is a class of evolutionary algorithm. This algorithm is suited for continuous values, usually with an elitist selection and a specific mutation (crossover is used rarely). In ES, the individual is made of the objective variables plus some other parameters guiding the search. Thus, a ES facilitates a kind of self-adaption by evolving the problem variables as well as the strategy parameters at the same time. Hence, the parameterization of an ES is highly customizable.

 1 t = 0
 2 initialize P(t)
 3 evaluate structures in P(t)
 4 while not end do
 5    t = t + 1
 6    select C(t) from P(t-1)
 7    recombine structures in C(t) forming C'(t)
 8    mutate structures in C'(t) forming C''(t)
 9    evaluate structures in C''(t)
10    replace P(t) from C''(t) and/or P(t-1)

The Evolution Strategy skeleton (ES) requires the classes:

• Problem
• Solution
  

The class Problem corresponds to the definition of a problem instance. The skeleton filler must provide a complete definition of this class.

And finally, the class Solution corresponds to the definition of a solution (feasible or not) of a problem instance. The skeleton filler must provide a complete definition of the class Solution.

In adition, the user must configure the following algorithm parameters (in file ES.cfg):

• number of independent runs.
• number of generations.
• size of population.
• size of offsprings in each generation.
• replace mode (if replaces parents for offsprings, or only offsprings may be new parents).
• Selection operators parameters (selection of parents).
• Intra operators parameters (crossover and mutation parameters).
• Inter operators (operators to apply between sub-populations) parameters: operator number, operator rate, number of individuals and selection of individual to send and replace.
• Parallel Configuracion: interval of generation to refresh global state, running mode (synchronized or asyncronized) and interval of generations to check solutions from other populations.

There are several basic steps to running a problem solve with ES skeleton:

1. Change to the problem directory

cd Mallba/rep/ES/problem

2. Compile skeleton.

make

3. Configure algorithm parameters (ES.cfg file)
4. Run problem:
       4.1 Sequential Version:

make SEQ
     or
MainSeq ES.cfg_path instance_path res_file_path

        4.2 Parallel Version:
                     4.2.1 Configure Config.cfg file.
                     4.2.2 Configure pgfileLan (or pgfileWan) : machines where we run the program.
                     4.2.3 Run

make LAN
     or
make WAN