Genex v1.0.1-beta Genex behaviour View Source
Genex makes it easy to write Evolutionary Algorithms in Elixir. The process of creating an algorithm in Genex can be thought of in three phases:
- Problem Definition
- Evolution Definition
- Algorithm Execution
The
Genexmodule contains the behaviour necessary to encode a problem the Genex Way. This module implements "Phase 1" of the three phase process used by Genex. Genex problems consist of a minimum of 3 and at most 6 functions. These functions: - Define your solution space (
genotype/0anddatatype/0). - Define your objective function (
fitness_function/1andweights/0). - Define your termination criteria (
terminate?/1).
Implementing a Problem
A basic Genex problem consists of: genotype/0, fitness_function/1, and terminate?/1. To define your first problem, create a new module with use Genex and declare each of these functions:
defmodule MyProblem do
use Genex
def genotype, do: Genotype.binary(10)
def fitness_function(chromosome), do: Enum.sum(chromosome.genes)
def terminate?(population), do: population.max_fitness == 10
endRunning a Problem
Genex injects a special function called run/1 into your problem definition. run/1 takes a number of optional configuration options you can use to change how your algorithm is executed (see configuration guide). To properly run your algorithm, open up iex and call MyProblem.run() or declare your module in .exs file and run mix run path/to/my_problem.exs.
Configuring Algorithms
Genex supports the following configuration options:
:title:Stringtitle of your algorithm. Defaults to"Genetic Algorithm".:population_size:Integersize of your initial population.:selection_type:Functionselection strategy. Defaults toGenex.Selection.natural().:crossover_type:Functioncrossover strategy. Defualts toGenex.Crossover.single_point().:mutation_type:Functionmutation strategy. Defaults to:none.:survival_selection_type:Functionsurvival selection strategy. Defaults toGenex.Selection.natural().:selection_rate:Floatbetween0and1orFunction. Defaults to0.8.:mutation_rate:Floatbetween0and1orFunction. Defaults to0.05.:survival_rate:Floatbetween0and1orFunction. Defaults to1 - selection_rate. All of these options are passed torun/1as aKeywordlist. See Configuring Genex for more.
Additional Utilities
The problem definition contains a few additional utilities that you might find useful to use. To find out more, see Additional Utilities.
Link to this section Summary
Callbacks
Function used to specify how genes are stored in the chromosome.
Genex supports any structure that implements Enumerable. After each generation, the algorithm "repairs" broke chromosomes by calling collection on each chromosome in the population. This is because many of the functions in Enum explicitly return a List.
This should be a reference to a function that accepts an Enum and returns an Enum. Most data structures come with a utility function for creating new versions from other types. An example is MapSet.new/1. Elixir's MapSet.new/1 accepts an Enum and returns a new MapSet.
The default version of this function returns a function that returns itself. In other words, the default version of this function simply returns the current representation of the chromosome. Because of how some of the functions in Enum work, this will always be a List.
Evaluates a chromosome.
The fitness function is your objective function(s). It is the function(s) you are trying to optimize. The fitness function will be used during an evolution to evaluate solutions against one another.
Genex supports multiple objectives. You can optimize multiple objectives at once by returning an Enum of objectives calculated from the chromosome.
It's important to note that chromosome is of type Chromosome.t. Typically you want to assess fitness based on the genes of the chromosome, so you must access the genes field of the struct.
Generates a random gene set.
This function is called n times, where n is the size of your population, at the beginning of the genetic algorithm to define the initial population. It's important that this function return random values, otherwise your initial population will consist of entirely the same set of solutions.
The genotype of your problem represents how solutions to problems are encoded. It's your solution space. Genex.Tools.Genotype implements several convenience functions for representing different types of solutions. See the Representing Solutions guide for information about the different ways to represent solutions with Genex.
Tests the population for some termination criteria.
The termination criteria is assessed at the beginning of every generation. If termination criteria is met, the algorithm halts and returns the current population.
Termination criteria assesses the entire population. The Population struct contains several convenience features you can use to determine when to halt your algorithms. You can stop on max fitness, generations, or some arbitrary feature determined from the population. Check out Crafting Termination Criteria to see more.
Returns initial weights associated with each objective.
The weights associated with each objective represent how "important" each weight is to your final solution. Typically, you want to weigh objectives equally, so your weights will consist of all ones.
The default implementation of this function returns 1. This represents a single objective maximization problem. To minimize the objective function, declare a weight of -1. You can also achieve this in the fitness function; however, declaring weights makes your algorithm more expressive and easier to follow.
Link to this section Callbacks
Function used to specify how genes are stored in the chromosome.
Genex supports any structure that implements Enumerable. After each generation, the algorithm "repairs" broke chromosomes by calling collection on each chromosome in the population. This is because many of the functions in Enum explicitly return a List.
This should be a reference to a function that accepts an Enum and returns an Enum. Most data structures come with a utility function for creating new versions from other types. An example is MapSet.new/1. Elixir's MapSet.new/1 accepts an Enum and returns a new MapSet.
The default version of this function returns a function that returns itself. In other words, the default version of this function simply returns the current representation of the chromosome. Because of how some of the functions in Enum work, this will always be a List.
fitness_function(chromosome)
View Sourcefitness_function(chromosome :: Genex.Types.Chromosome.t()) :: number() | Enum.t()
Evaluates a chromosome.
The fitness function is your objective function(s). It is the function(s) you are trying to optimize. The fitness function will be used during an evolution to evaluate solutions against one another.
Genex supports multiple objectives. You can optimize multiple objectives at once by returning an Enum of objectives calculated from the chromosome.
It's important to note that chromosome is of type Chromosome.t. Typically you want to assess fitness based on the genes of the chromosome, so you must access the genes field of the struct.
Generates a random gene set.
This function is called n times, where n is the size of your population, at the beginning of the genetic algorithm to define the initial population. It's important that this function return random values, otherwise your initial population will consist of entirely the same set of solutions.
The genotype of your problem represents how solutions to problems are encoded. It's your solution space. Genex.Tools.Genotype implements several convenience functions for representing different types of solutions. See the Representing Solutions guide for information about the different ways to represent solutions with Genex.
terminate?(population)
View Sourceterminate?(population :: Genex.Types.Population.t()) :: boolean()
Tests the population for some termination criteria.
The termination criteria is assessed at the beginning of every generation. If termination criteria is met, the algorithm halts and returns the current population.
Termination criteria assesses the entire population. The Population struct contains several convenience features you can use to determine when to halt your algorithms. You can stop on max fitness, generations, or some arbitrary feature determined from the population. Check out Crafting Termination Criteria to see more.
Returns initial weights associated with each objective.
The weights associated with each objective represent how "important" each weight is to your final solution. Typically, you want to weigh objectives equally, so your weights will consist of all ones.
The default implementation of this function returns 1. This represents a single objective maximization problem. To minimize the objective function, declare a weight of -1. You can also achieve this in the fitness function; however, declaring weights makes your algorithm more expressive and easier to follow.