Optimization with Genetic Algorithm - Part 1

#### I’m dividing this post into 2 parts:

Member of the Evolutionary Computing set of algorithms, a Genetic Algorithm is a metaheuristic technique (first proposed by John Holland in the early 1970’s) based on Darwin’s theory of evolution (which defines the idea of natural selection observed in earth’s nature) that tries to gradually improve an objective function value. The analogy is that a problem’s answers can be used to build a new generation of improved answers for the same problem the same way a population of individuals can be used to build a new generation of improved individuals.

A really nice description for this intuition is given by Marek Obitko at his “Introduction to Genetic Algorithms” website (which is a highly recommend reading):

A Genetic Algorithm starts “with a set of solutions (represented by chromosomes) called population. Solutions from one population are taken and used to form a new population. This is motivated by a hope, that the new population will be better than the old one. Solutions which are selected to form new solutions (offspring) are selected according to their fitness - the more suitable they are the more chances they have to reproduce. This is repeated until some condition (for example number of populations or improvement of the best solution) is satisfied.”

This intuition is usually translated into the following procedure:

1. Create current chromosome generation (a list of feasible answers to the problem at hand)
2. While condition is not meet (a specific fitness values or the max number of generations)
1. While new generation is not full
1. Selection (select parents based on fitness (not necessarily a pair), the higher the fitness higher are the chances of being selected)
2. Crossover (cross them based on probability)
3. Mutate (mutation probability is usually low)
2. Set new generation as current
3. Elitism (the fittest of the previous generation can be included in the new one)
3. Return the overall fittest individual

For all the sections below, please consider reading the “Introduction to Genetic Algorithms” website for more detail. I’m adding them here mostly for quick reference.

## Definitions

• Generation: Children of a crossover operation
• Chromosome: A set of genes; usually a feasible answer to a problem.
• Population: Number of chromosomes in the Generation
• Gene: A single information of the chromosome; usually a bit or a single characteristic of the feasible answer.
• Genotype: A particular set of genes and settings that determines the phenotype.
• Phenotype: The physical expression of the genotype - the organism itself.

## Selection

• Roulette Wheel Selection
Calculate the sum of fitnesses S from current generation and generate a random number R from 0 to S. Iterate through the current generation while accumulating each fitnesses A, return first chromosome where A is greater than R. This selection strategy greatly benefits higher fitness chromosome.
• Rank Selection
Sort the current generation by ascending fitness. Number them from 1 to N and execute the Roulette Wheel Selection using their new number instead of their fitnesses. This changes the probability distribution on the Roulette Wheel Selection to be more uniform and less focused on higher fitness chromosomes.
Generate new chromosomes with the fittest parents, then copy the old generation, remove low fitness chromosomes and add the new chromosomes.

## Crossover

• Binary Encoding
A chromosome is represented by a string of sequential bits. eg. 10100101
• Permutation Encoding
A chromosome is represented by a string of sequential numbers or letters. eg. ABDEFGCH or 93847565
• Value Encoding
A chromosome is represented by a string of real numbers, words or any other multicharacter value. eg. (word1)(word3)(word4)(word5) or (21.5)(32.4)(1.45678)

## Mutation

Based on a mutation parameter, just randomly change or switch one or more genes. eg. ABCDEFGH -> ABFDECGH

## Time Complexity

If the main loop will run until a specific low fitness chromosome is reached, then it’s difficult to predict when it will stop or if it will stop at all. That’s one of the reasons why a maximum number of generations if given to the final algorithm, to ensure it will stop and to better balance it’s speed/quality. In this case time complexity is mostly given by the strategy defined for Crossover since the Population size, number of Generations, Selection iterations and Mutation iterations will be a constant number, unrelated to the input size.

## Application

Jump to Part 2 (soon) for a hands-on algorithm explaining how to implement a Genetic Algorithm to solve the Traveling Salesman Problem.