dtc-machiene-learning

A light weigh JS utility for basic and quick ML problems produced by DownToCrypto

Motovation

I was looking to create a simple ans easy to use utility for AI to aid in automated trading.

Tech

Node JS palne and simple

Features

At this point it is still in its infancy. As such it is simply an easy to use genetic optomization tool. More NL and deep learning is to come.

Examples

Genetic Optomization

Gene Types

• GeneTypes.int: any real integer value
• GeneTypes.float: any real float value
• GeneTypes.bool: true or false

GeneTypes.int and GeneTypes.float have a default min and max of +/-10

Mutation Types

• MutationTypes.uniform: sets the gene to a random number between the min and max inclusivly
• MutationTypes.boundry: sets the gene to the max or min at random
• MutationTypes.percent: changes the value to within +/- the specafied percent of the current value at random

Putting it all together

Alaways start with

const {
  Population,
  MutationTypes,
  GeneTypes,
} = require("dtc-ml").Genetic;

Making the Organisms

const buildingblocks = [
{ type: GeneTypes.int, min: 0, max: 10 },
{ type: GeneTypes.int, min: 0, max: 10 },
];

Making the fitness test for the population

function FitnessTest(individual) {
  return individual[0].value * individual[1].value;
}

Making a population of 100 organisms out of the building blocks

const populationSize = 100;
let population = new Population(populationSize, buildingblocks, FitnessTest);

Randomizes all of the genes for a population

population.randomize();

Running the fitness test and score all individuals

population.runFitnessTests();

Cache any individuals to save time down the road. This is optional

population.saveFamilyTree();

Determin breeding pool based off individual scores

population.selection();

Breed next generation and determine what percentage of the top performers carry over to the next generation

population.breed(0.05);//top 5% stay till next generation

Mutate the population based on the selected method and percentage rate

population.mutate(MutationTypes.uniform, 0.05);

The "transitionToNextGeneration" method rolls selection, breed and mutate into 1 call.

population.transitionToNextGeneration(0, MutationTypes.uniform, 0.05);

Here is it all together with some periferals to record the findings

const {
  Population,
  MutationTypes,
  GeneTypes,
} = require("dtc-ml");

const buildingblocks = [
  { type: GeneTypes.int, min: 0, max: 10 },
  { type: GeneTypes.int, min: 0, max: 10 },
];

function FitnessTest(individual) {
  return individual[0].value * individual[1].value;
}

const populationSize = 100;

let population = new Population(populationSize, buildingblocks, FitnessTest);

population.randomize();

const generations = 10;

for (let i = 1; i <= generations; i++) {
  population.runFitnessTests();
  population.saveFamilyTree();
  if (i !== generations) {
    population.transitionToNextGeneration(0.1, MutationTypes.uniform, 0.05);
  }
}

console.log(population.getFittest());

The outshould be the below. Note there is some randomness involved so you may get a gene that has a value of 9. If you do just run it again.

Individual(2) [
  Int { min: 0, max: 10, value: 10 },
  Int { min: 0, max: 10, value: 10 },
  fitness: 100,
  id: 46724426
]

Coming Soon

• Built in exit conditions for genetic learning
• "runGenerations" method to contain for loop internally to the population class
• Neurons