Skip to content

An Evolutionary Algorithm written in JavaScript that can take in any generic module representing a problem to be optimized

Notifications You must be signed in to change notification settings

methodswithclass/evolve

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

78 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Evolutionary Algorithm

This library optimizes a strategy to solve a problem by way of an evolutionary algorithm, a model of biological evolution: rank fitness, crossover of best performing, mutation, run next generation.

The initial generation is made up of individuals with random strategies, the output is a strategy that is optimized to solve the given problem

The algorithm is agnostic to the actual problem, it is primarily specified through the getFitness function

Usage

npm install @methodswithclass/evolve

import Evolve from @methodswithclass/evolve

The input consists of options for the evolutionary algorithm and the specification of your problem.

Mutate

called when processing each individual

returns a brand new strategy or a mutated strategy

required, can be async

strategy is the central object that is being optimized (any type)

if input strategy is undefined then return a brand new strategy with random values, if it exists, return a copy of it, with some values newly randomized by some percentage

this function would typically return a randomly valued item (within a relevant range)

const mutate = async (strategy) => {
	return newStrategy // can be any type, typically the strategy is a consistent type
}

BeforeEach

called before processing each new generation

returns a boolean on whether to process the next generation, the current generation will finish

optional, can be async

generation is the number of the current generation

const beforeEach = async (input) => {
	const { generation, first, last } = input;
	return true/false // must explicitly return false if you want the process to stop
}

AfterEach

called after processing each generation

returns nothing

optional, can be async

generation is the number of the current generation

best is the top performing individual of the generation: { strategy, fitness }

const afterEach = async (input) => {
	const { generation, best } = input;
	no return
}

OnEnd

called when all the generations have been processed or evolution told to stop (beforeEach returns false)

optional, can be async

generation is the number of the current generation

best is the top performing individual of the generation: { strategy, fitness }

const onEnd = async (input) => {
	const { generation, best } = input;
	no return
}

GetFitness

the primary definition of your problem, it takes a strategy and runs it against the problem to produce a measure of its performance, the fitness

required, can be async

populations for every generation run all individuals simultaneously, indexes not guaranteed to be in order

strategy is the object (any type) formed by mutate(), fitness is calculated by some process on this object

index: {generation number}#{index of individual in population array}

id: {unique id for generation}#{unique id for individual}

returns the fitness of an individual (any type), can be async

this fitness is then passed into rank

const getFitness = async (input) => {
	const { index, id, strategy } = input;
	return fitness;
}

Combine

called during generation crossover

required, can be async

strategies are pulled from the top performers of the generation at random

const combine = async (strategyA, strategyB) => {
	return newStrategy
}

Rank

comparator according to Array.sort(),

optional, must be synchronous

some function according to your optimization and fitness scheme/type

if not included Evolve will treat the fitness as a number, and will maximize it

const rank = (fitnessA, fitnessB) => {
	return fitnessB - fitnessA;
}

Initialization

const input = {
	first, // first generation number, required
	last, // last generation number, must be greater than first, required
	best, // initial best performer { strategy, fitness }, optional, used when first is not 1, when this is included, all individuals in the first generation are mutated versions of this
	popTotal, // number of individuals per generation, required, defaults to 100
	beforeEach,
	afterEach,
	onEnd,
	getFitness,
	combine,
	rank,
	mutate,
}

const evolve = Evolve(input);

await evolve.start();

evolve.stop();

process auto-stops when the current generation equals the value in last

Example

const totalGenes = 100;

const getGene = () => {
	return Math.random();
};

const mutate = (strategy) => {
	const newStrategy = [];

	if (!strategy) {
		for (let i = 0; i < totalGenes; i++) {
			newStrategy.push(getGene());
		}
		return newStrategy;
	}

	const mutatedStrategy = strategy.map((item) => {
		if (Math.random() < 0.02) {
			return getGene();
		}

		return item;
	});

	return mutatedStrategy;
};

const combine = (a, b) => {
	const aSegment = a.slice(0, totalGenes / 2);
	const bSegment = b.slice(totalGenes / 2);

	return [...aSegment, ...bSegment];
};

const getFitness = ({ strategy }) => {
	const fitness = strategy.reduce((total, item) => {
		if (1 - item < 0.01) {
			return total + 50;
		}
		return total - 20;
	}, 0);

	return fitness;
};

let bestPerformer;

const evolve = Evolve({
	first: 1,
	last: 100,
	popTotal: 100,
	afterEach: ({ generation, best }) => {
		console.log('current generation', generation);
		bestPerformer = best;
	},
	getFitness,
	mutate,
	combine,
});

await evolve.start();

const { strategy, fitness } = bestPerformer;

this resulting strategy will be an array of values all close to 1;