themladypan / pyga

python genetic algorithm implementation

pyga

Python implementation of Genetic Algorithm

Installation

pip3 install pygenal

Usage

this test is able to find global maximum of 6th order polynom with two variables in 500ms with precision approximately 10^-4 to 10^-5 % (depending on hardware)

#!/usr/bin/env python3

# import required classes
from pygenal.ga import Population, Individual, Gene, geneTypes, Duration
import numpy as np
import time


# create population of your own, override fitness method to suit your needs.
# your system evaluation goes here, for sake of demonstration, use polynom
class Polynom(Population):
    def fitness(self, individual):
        x = individual.get("x").value
        y = individual.get("y").value
        color = individual.get("color").value

        # test against function with global maximum:
        # max{x + 3 x^2 - x^4 - y^6 + y/3 - y^3 4 - y - 5} ≈ 3.35864
        # at (x, y) ≈ (1.30084, -1.27413)
        # + favorize chocolate over shorter colors, just for demonstration
        return (x + 3*(x**2) - x**4 - y**6 + y/3.0 - (y**3)*4 - y - 5) + len(color)/10


# introduce non-number options such as color, good for selections
colors = ["brown", "green", "grey", "blue", "chocolate"]


if __name__ == '__main__':

    # spawn individual
    i1 = Individual()

    # construct "DNA"
    i1 += Gene("x", geneTypes.REAL, np.random.randint(-10, 10), min=-10, max=10)
    i1 += Gene("y", geneTypes.REAL, np.random.randint(-10, 10), min=-10, max=10)
    i1 += Gene("color", geneTypes.VALUE, np.random.choice(colors), availableOptions=colors, dominant=True)

    # create tribe with size of 100 individuals based on your first Individual
    population = Polynom(
            species=i1,
            size=100,
            chanceOfMutationStart=.5,
            chanceOfMutationStop=0.0001
        )

    tStart = time.time()

    # start evolution until:
    #   you didn't pass 1000 generations OR
    #   fittest didn't change for 200 generations OR
    #   2.5s didn't passes yet
    population.evolve(
            allowCrossover=True,
            generations=100,
            verbose=True,
            timeout=Duration(seconds=0, miliseconds=500),
        )
    print(f"Evolved in {time.time()-tStart}s, precision: {(4.258642115713602 - population.fittest.score)*100}%")
    print(f"Fittest: {population.fittest}, genes: {repr(population.fittest)}")