js-synaptic-population

A population of neural networks built over (synapticjs)[https://caza.la/synaptic/] using evolution process instead of backpropagation

Installation

npm i synaptic-population

Usage

import Population from 'synaptic-population';        // es6 import

or

const Population = require('synaptic-population');   // classic nodejs require

// Create a population with some optionals rules
const population = new Population({
  demography : 10,            // Number of brains in the population. Default 10.
  eliteDemography : 4,        // Number of brains that will survive natural selection on each generation. Default 4.
  extinctionFitness : null,   // If all brains in a generation get a lower score that this parameter, then all the population is reset. Default null.
  inputs : 1,                 // Number of inputs required to activate brains. Default 2.
  outputs : 1,                // Number of outputs returned from the brains. Default 1.
  hiddenLayers : [8],         // Array of numbers representing the hidden layers neurons. Default [8] (one hidden layer of 8 neurons)
  mutateRate : 0.2,            // Mutation rate from 0 (no mutation) to 1 (full mutation) applied when evolving
  trainedPop: null            // Retrieve a previously exported population (see method toJSON())
});

// Activate brains
population.activateBrain(brainID, inputs);

// Set fitness of brains
population.setBrainFitness(brainID, fitness);

// Evolve
population.evolve();

// And do it again until your population does what you want them to do !

See x

API

Properties

• brains : array READ-ONLY // Array of neural networks.
• demography : int READ-ONLY, // Number of brains in the population. Default 10 .
• eliteDemography : int READ-ONLY, // Number of brains that will survive natural selection on each generation. Default 4.
• extinctionFitness : number or null READ-ONLY, // If all brains in a generation get a lower score that this parameter, then all the population is reset. Default null.
• generation : int READ-ONLY // Number of generation that occurred into the population.
• hiddenLayers : array of int READ-ONLY, // Array of numbers representing the hidden layers neurons. Default [8] (one hidden layer of 8 neurons)
• inputs : int READ-ONLY, // Number of inputs required to activate brains. Default 1.
• mutateRate : float [0-1] // Mutation rate from 0 (no mutation) to 1 (all the children mutates) applied when evolving. Default 0.2.
• mutateFactor : int // The degree of mutation applied when mutating a child. Default 3.
• outputs : int READ-ONLY, // Number of outputs returned from the brains. Default 1.
• trainedPop: JSON object READ-ONLY // retrieve a previously exported population (see method toJSON()). Default null.

Methods

// Create a population with some optionals rules

• constructor({

  • demography = 10, // Number of brains in the population. Default 10.
  • eliteDemography = 4, // Number of brains that will survive natural selection on each generation. Default 4.
  • extinctionFitness = null, // If all brains in a generation get a lower score that this parameter, then all the population is reset. Default null.
  • inputs = 2, // Number of inputs required to activate brains. Default 2.
  • outputs = 1, // Number of outputs returned from the brains. Default 1.
  • hiddenLayers = [8], // Array of numbers representing the hidden layers neurons. Default [8] (one hidden layer of 8 neurons)
  • mutateRate = 0.2, // Mutation rate from 0 (no mutation) to 1 (full mutation) applied when evolving. Default 0.2
  • trainedPop = null // Retrieve a previously exported population (see method toJSON()). Default null

  });

// Activate a brain neurons with inputs array

• activateBrain(id, inputs) id : the brain (individual) ID in your population inputs : an inputs array for activating neurons

// Set the fitness of a brain

• setBrainFitness(id, fitness) id : the brain (individual) ID in your population fitness : the score to assign to the brain

// Returns a brain given its ID

• getBrain(id) id : the brain (individual) ID in your population

// Keep the bests, kill the rest and replace them with children of the bests

• evolve()

// Reset the population to generation 0, all training is lost

• reset()

// Export the current population into JSON object

• toJSON()