chapter five

5 Advanced evolutionary approaches

 

In this chapter

  • Considering options for the steps in the genetic algorithm life cycle
  • Adjusting a genetic algorithm to solve different problems
  • Configuringadvanced parameters for a genetic algorithm life cycle

Evolutionary algorithm life cycle

The general life cycle of a genetic algorithm is outlined in chapter 4, which is a prerequisite for this chapter. Here, we consider other problems that a genetic algorithm might be well-suited to solve, why some of the approaches demonstrated so far won’t work, and what alternatives we might try. As a reminder, the general life cycle of a genetic algorithm is

  1. Creating a population. This step involves generating an initial set of random candidate solutions.
  2. Measuring the fitness of individuals in the population. This step determines how good a specific solution is by using a fitness function that scores solutions to determine how good they are.
  3. Selecting parents based on their fitness. This step involves selecting parents that will produce offspring.
  4. Reproducing individuals from parents. This step generates new candidates by combining the genes of selected parents and introducing random mutations to the offspring.
  5. Populating the next generation. This step involves selecting individuals and offspring from the population that will survive to the next generation.

Alternative selection strategies

Rank selection: Even the playing field

Tournament selection: Let them fight

Elitism selection: Choose only the best

Real-value encoding: Working with real numbers

Real-value encoding at its core

Arithmetic crossover: Reproduce with math

Boundary mutation

Arithmetic mutation

Importance of the fitness function