Genetics

A genetic algorithm library for Swift.

• Features
• Installation
• Usage
• License

Features

Selection

• Rank
• Roulette
• Tournament
• Stochastic
• Custom

Crossover

• N-Point
• Uniform
• Custom

Installation

Swift Package Manager

Add the following to the dependencies in your Package.swift file.

dependencies: [
  .package(url: "https://github.com/colinc86/Genetics.git, from: "1.0.0")
]

Manually

1. Navigate to your project's directory and clone the Genetics repository.

$ git clone https://github.com/colinc86/Genetics.git

2. Add Genetics.xcodeproj to your Xcode project.
3. Add the Genetics framework to your project.

Usage

For a working example, see the GeneticsExample.

Preparing for Evolution

A few things need to be set up before evolution can take place. We must have a population of chromosomes to evolve, and a set of parameters to guide the evolution of each generation. The Chromosome, Population and EvolverConfiguration objects are what we'll use.

Generating a Population of Chromosomes

The first thing you'll need to do is generate a population of chromosomes. The easiest way to do that is by using the static function Population.generate(populationSize:chromosomeLength:generatingFunction:). The following example generates a Population of 4 chromosomes with length 2 by using the provided generating function.

var population = Population.generate(populationSize: 4, chromosomeLength: 2) { (_, _) -> Double in
  return arc4random_uniform(20)
}

You can also create chromosomes manually by initializing them with an array or an array literal.

let chromosomeA = Chromosome([1.0, 2.0, 3.0])
let chromosomeB = Chromosome(1.0, 2.0, 3.0)

Then, initialize a Population with an array of chromosomes.

let population = Population([chromosomeA, chromosomeB])

Creating an EvolverConfiguration

An instance of EvolverConfiguration contains parameters for use by an Evolver when performing evolution. Creating one is easy:

let configuration = EvolverConfiguration(selectionMethod: .rank, crossoverMethod: .point(count: 1), elitism: .none, crossoverRate: 0.1, mutationRate: 0.1)

Evolution

The class responsible for evolving a population of chromosomes is the Evolver. You create an evolver by giving it a configuration, fitness function, and a mutation function. The fitness function is responsible for returning the fitness of a chromosome in the population, and the mutation function is responsible for mutating a single element of a chromosome.

Creating an Evolver

The following example creates an instance of Evolver with configuration and provides fitness and mutation functions. The fitness function divides 10.0 by the sum of all of the elements in the chromosome. The mutation function randomly adds or subtracts 1 from an element of the chromosome and returns the result. This will result in the fittest chromosomes having elements whos sum is near 10.0.

let evolver = Evolver(configuration: configuration, fitnessFunction: { (chromosome: Chromosome) -> Double in
  // Fitness function
  return 10.0 / chromosome.reduce(0, +)
}, mutationFunction: { (chromosome: Chromosome, index: Int) -> Double in
  // Mutation function
  var value = chromosome[index]  
  value += arc4random_uniform(2) == 0 ? -1.0 : 1.0
  
  return value
})

Evolving a Population

The easiest way to evolve a population is by using the evolver's evolve(population:shouldContinue:) function.

do {
  try evolver.evolve(population: &population, shouldContinue: { (config: inout EvolverConfiguration, pop: Population) -> Bool in
    return pop.generation < 100
  })
}
catch let error {
  print("\(error)")
}

The preceding example evolves the population 100 times. Notice that the shouldContinue closure takes an inout EvolverConfiguration parameter. The evolver gives you the chance to modify its configuration before the next evolution cycle.

You can also call evolve once without providing the shouldContinue closure.

do {
  for _ in 0 ..< 100 {
    try evolver.evolve(population: &population)
  }
}
catch let error {
  print("\(error)")
}

License

Genetics is released under the MIT license.

See LICENSE.