intelligence
Version:
Machine learning library written in javascript
242 lines (225 loc) • 8.3 kB
JavaScript
var events = require('events');
var utils = require('./../infrastructure/utils');
var selectionStrategies = require('./selectionStrategies');
/**
* Genetic algorithm population
* @constructor
* @param {object} options - Population options
* @param {Individual} options.baseIndividual -
* @property {object} options - Population options
*/
var Population = function (options) {
this.options = options;
this.individuals = null;
this.validateRequiredOptions();
this.setDefaultOptionsIfNotProvided();
this.initialise();
events.EventEmitter.call(this);
return this;
};
utils.inherits(Population, events.EventEmitter);
/**
* Throws an exception if a required option is missing
* @throws An exception is thrown if a required option is missing
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.validateRequiredOptions = function () {
if (!this.options) {
throw "options are required";
} else if (!this.options.baseIndividual) {
throw "option 'baseIndividual' is required";
} else if (!this.options.crossoverStrategy) {
throw "option 'crossoverStrategy' is required";
} else if (!this.options.fitnessFunction) {
throw "options 'fitnessFunction' is required";
}
return this;
};
/**
* Sets default values for options that have not been defined
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.setDefaultOptionsIfNotProvided = function () {
if (!this.options.populationSize) {
this.options.populationSize = 100;
}
if (this.options.crossoverRate === undefined) {
this.options.crossoverRate = 0.75;
}
if (this.options.mutationRate === undefined) {
this.options.mutationRate = 0.2;
}
if (!this.options.tournamentSize) {
this.options.tournamentSize = Math.ceil(this.options.populationSize * 0.05);
}
if (!this.options.selectionStrategy) {
this.options.selectionStrategy = selectionStrategies.tournament;
}
return this;
};
/**
* Creates a randomly generated population of individuals
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.initialise = function () {
this.individuals = [];
for (var i = 0; i < this.options.populationSize; i++) {
this.individuals.push(this.options.baseIndividual.createNew());
}
return this;
};
/**
* Calculates the fitness of each individual where the fitness value is null
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.evaluateFitness = function () {
for (var i = 0; i < this.individuals.length; i++) {
var individual = this.individuals[i];
if (individual.fitness === null) {
individual.fitness = this.options.fitnessFunction(individual);
}
}
return this.filterNanFitness();
};
/**
* Performs crossover using the crossoverStrategy function defined in the population options
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.crossover = function () {
this.evaluateFitness();
var limbo = [];
if (this.options.elitism) {
var elite = this.getFittestIndividuals(this.options.elitism);
for (var i = 0; i < elite.length; i++) {
limbo.push(elite[i].copy());
}
}
while (limbo.length < this.individuals.length) {
var selections = this.options.selectionStrategy(this.individuals, this.options);
if (utils.random() < this.options.crossoverRate) {
var elections = this.options.crossoverStrategy(selections, this.options);
selections = elections;
for (var i = 0; i < selections.length; i++) selections[i].fitness = null;
}
limbo = limbo.concat(selections);
}
this.individuals = limbo;
return this;
};
/**
* Mutates the population based on the mutationRate property in the population options
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.mutate = function () {
this.evaluateFitness();
var elite = this.options.elitism ? this.getFittestIndividuals(this.options.elitism) : null;
for (var i = 0; i < this.individuals.length; i++) {
if (!elite || elite.indexOf(this.individuals[i]) === -1) {
if (utils.random() < this.options.mutationRate) {
this.individuals[i].mutate();
this.individuals[i].fitness = null;
}
}
}
return this;
};
/**
* Returns a specified number of individuals with the best fitness rating in the population
* @param {number} [numIndividuals=1] - The number of individuals to return
* @returns {Individual[]} An array of the fittest individuals in the population
*/
Population.prototype.getFittestIndividuals = function (numIndividuals) {
this.evaluateFitness();
var self = this;
this.evaluateFitness();
if (!numIndividuals) numIndividuals = 1;
return this.individuals.sort(function (a, b) {
if (a.fitness === null && b.fitness === null) {
return 0;
} else if (a.fitness === null) {
return -1;
} else if (b.fitness === null) {
return 1;
} else {
return self.options.isMinimise ? a.fitness - b.fitness : b.fitness - a.fitness;
}
}).slice(0, numIndividuals);
};
/**
* Calculate the average fitness of all individuals in the population (exluding infinite values)
* @returns {number} Average fitness of the population
*/
Population.prototype.getAverageFitness = function () {
this.evaluateFitness();
var sum = 0;
for (var i = 0; i < this.individuals.length; i++) {
if (isFinite(this.individuals[i].fitness)) {
sum += this.individuals[i].fitness;
}
};
return sum / this.individuals.length;
};
/**
* Applies a single iteration of crossover and mutation to the population
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.step = function () {
return this.evaluateFitness().crossover().mutate().evaluateFitness();
};
/**
* Trains the population over a specified number of generations
* @param {number} numGenerations - The number of generations to train the population over
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.train = function (numGenerations) {
if (numGenerations <= 0) {
throw "'numGenerations' must greater than 0";
} else {
for (var i = 0; i < numGenerations; i++) {
this.step();
this.emit('generationCompleted', this, i);
}
}
this.emit('trainingCompleted', this);
return this;
};
/**
* Subsitutes an individuals fitness to positive or negative infinity if it isNaN
* @returns {Population} Reference to current object for chaining
*/
Population.prototype.filterNanFitness = function () {
var value = this.options.isMinimise ? Number.POSITIVE_INFINITY : Number.NEGATIVE_INFINITY;
for (var i = 0; i < this.individuals.length; i++) {
var individual = this.individuals[i];
if (isNaN(individual.fitness)) {
individual.fitness = value
}
}
return this;
};
/**
* Save the population to a file
* @param {string} filePath - Path to file
* @callback {writeToFileCallback} cb - Callback handler
*/
Population.prototype.saveToFile = function (filePath, cb) {
var serialised = utils.serialise(this);
utils.writeToFile(filePath, serialised, cb);
};
/**
* Load a population from a file
* @static
* @param {string} filePath - Path to file
* @callback {loadFromFileCallback} cb - Callback handler
*/
Population.loadFromFile = function (filePath, cb) {
utils.readFromFile(filePath, function (err, data) {
if (err) {
cb(err);
} else {
var deserialised = utils.deserialise(data);
return deserialised;
}
});
};
exports.Population = Population;