Genetix

A simple library to test genetic algorithms

Installation

npm install -S genetix

Description

With Genetix you can create simple genetic algorithms. Its flexibility opens a lot of doors, and lets you customize every aspect of the algorithm.

Keep in mind that this is a library meant for testing and not a library meant for actual production/calculus code. It is SLOW and unoptimized.

Genetix uses 3 main classes Gene, Genome and Population.

tl;dr

To use Genetix you first initialize a Population object with a fitness function and generators. generators is a list of functions that return Genes.

Example:

let Genetics = require('genetix');
 
// Create generator function
let generator = function(current_genes){
    let int = Math.floor(Math.random() * 30) + 1;
    return new Genetics.Gene(int);
}
 
// All generators are the same
let generators = new Array(4).fill(generator);
 
// Create population with generators and fitness function
let pop = new Genetics.Population(generators, function(genes){
    // retrieve gene values
    let vals = genes.map(e => e.value());
 
    // return inverse of 30 - function (so that fitness is an increasing curve)
    let inverse = 30 - (vals[0] + 2*vals[1] + 3*vals[2] + 4*vals[3]);
    return 1/inverse;
});
 
// Evolve until a satisfying genome is found (30 is the target, 1000 max generation)
let last_gen = pop.auto(function(genes){
    let vals = genes.map(e => e.value());
    return (vals[0] + 2*vals[1] + 3*vals[2] + 4*vals[3]);
}, 30, 1000);
 

fitness takes as argument the genes of every Genome and expects a number as output.

Performance

You should read this at the end, but I know you won't read the whole doc, so I'm putting it here.

You should test many different mutation_rates (between 0 and 1 in probability) to find optimal execution time.

As an example, when running examples/mutation_rate.js, I get 1min50 sec on average and no exact match for mutation_rate = 0.05 and 400ms for mutation_rate=0.2 with an exact match (running 2.3GHz i7);

Try running the example on your computer.

Concepts

Population

A population is like a generation (actually, a generation in this case is composed of a population). It contains individuals, with a maximum number of them, capable of reproduction.

These indiivuals reproduce to combine their genes and go towards a given goal, measured by the fitness function.

The population class is defined as

Constructor

new Population(generators[], fitness(), [crossover()], [params], [population])

Param	Required	Type	Description
generators	true	Array of functions	Generators is an array of functions that return new Gene(val), they take current_genes[] (with Genes) as argument
fitness	true	Fitness function	Returns number, takes array of Genes as only argument
crossover	false	Crossover function	Returns list of new Genes, takes (parent1_genes, parent2_genes, generators, mutation_rate)
params	false	Params Object	Contains {mutation_rate:0.05, maxpop:100}

Methods

.fit()

Executes fitting of all Genomes in population

.breed(was_fit)

Breeds and returns new pre-filled Population. was_fit is a boolean that specifies if the Population was already fit, or needs fitting. Default is false;

.auto(pheno(), target, [max_generations], [gen_callback)

Automatically breeds populations until one reaches the given target.

Param	Required	Type	Description
pheno	true	Phenotype function or false	Function that takes list of Genes and returns some comparable/human-redable value
target	true	Comparable value (number/string) or Object	Used to compare population Phenotypes or object {target: <VAL>, compare: function(phenotype){return <bool>}}
max_generations	false	Number	Maximum number of generations before giving up. Default is 1000
gen_callback	false	Function	Called every time a new generation is bred. Takes {current_population, max_fitness, generation_number} as only argument

Genome

A Genome is practially the same thing as an Individual. It is composed of a list of Genes.

Constructor

new Genome(genes=[], fitness, crossover, mutation)

Methods

.fit()

Fits the Genome using fitness function given in constructor or Population.

.crossover(genome2, generators)

Crosses over curernt Genome with given Genome using given crossover in constructor or Population. Using mutation_rate randomly decides to add a random mutation to genes using given generators.

.pheno(fct)

Renders phenotype using given Pheno function.

Gene

A gene is the basic component of a Genome.

Constructor

new Gene(value, type=null)

Param	Required	Type	Description
value	true	Value or Function	Value used by the gene or returned by the gene when needed
type	false	String	'dominant' or 'recessive'

Methods

.value()

Returns value for gene. If function, executes and returns result.