Module `circle_evolution.evolution`

Responsible for evolving a specie to look like target image.

Expand source code

"""Responsible for evolving a specie to look like target image."""

import random

import numpy as np

from circle_evolution.species import Specie

import circle_evolution.fitness as fitness


class Evolution:
    """Logic for a Species Evolution.

    Use the Evolution class when you want to train a Specie to look like
    a target image.

    Attributes:
        size (tuple): tuple containing height and width of target image (h, w).
        target (np.ndarray): target image for evolution.
        genes (int): the amount of circle to train the target image on.
        generation (int): amount of generations Evolution class has trained.
        specie (species.Specie): the Specie that is getting trained.
    """

    def __init__(self, size, target, genes=5):
        """Initializes Evolution class.

        Args:
            size (tuple): tuple containing height and width of target image (h, w).
            target (np.ndarray): target image for evolution.
            genes (int): the amount of circle to train the target image on.
        """
        self.size = size  # Tuple (y, x)
        self.target = target  # Target Image
        self.generation = 1
        self.genes = genes

        self.specie = Specie(size=self.size, genes=genes)

    def mutate(self, specie):
        """Mutates specie for evolution.

        Args:
            specie (species.Specie): Specie to mutate.

        Returns:
            New Specie class, that has been mutated.
        """
        new_specie = Specie(size=self.size, genotype=np.array(specie.genotype))

        # Randomization for Evolution
        y = random.randint(0, self.genes - 1)
        change = random.randint(0, 6)

        if change >= 6:
            change -= 1
            i, j = y, random.randint(0, self.genes - 1)
            i, j, s = (i, j, -1) if i < j else (j, i, 1)
            new_specie.genotype[i : j + 1] = np.roll(new_specie.genotype[i : j + 1], shift=s, axis=0)
            y = j

        selection = np.random.choice(5, size=change, replace=False)

        if random.random() < 0.25:
            new_specie.genotype[y, selection] = np.random.rand(len(selection))
        else:
            new_specie.genotype[y, selection] += (np.random.rand(len(selection)) - 0.5) / 3
            new_specie.genotype[y, selection] = np.clip(new_specie.genotype[y, selection], 0, 1)

        return new_specie

    def print_progress(self, fit):
        """Prints progress of Evolution.

        Args:
            fit (float): fitness value of specie.
        """
        print("GEN {}, FIT {:.8f}".format(self.generation, fit))

    def evolve(self, fitness=fitness.MSEFitness, max_generation=100000):
        """Genetic Algorithm for evolution.

        Call this function to begin evolving a Specie.

        Args:
            fitness (fitness.Fitness): fitness class to score species preformance.
            max_generation (int): amount of generations to train for.
        """
        fitness = fitness(self.target)

        self.specie.render()
        fit = fitness.score(self.specie.phenotype)

        for i in range(max_generation):
            self.generation = i + 1

            mutated = self.mutate(self.specie)
            mutated.render()
            newfit = fitness.score(mutated.phenotype)

            if newfit > fit:
                fit = newfit
                self.specie = mutated
                self.print_progress(newfit)

Classes

class Evolution (size, target, genes=5)

Logic for a Species Evolution.

Use the Evolution class when you want to train a Specie to look like a target image.

Attributes

size : tuple: tuple containing height and width of target image (h, w).
target : np.ndarray: target image for evolution.
genes : int: the amount of circle to train the target image on.
generation : int: amount of generations Evolution class has trained.
specie : species.Specie: the Specie that is getting trained.

Initializes Evolution class.

Args

size : tuple: tuple containing height and width of target image (h, w).
target : np.ndarray: target image for evolution.
genes : int: the amount of circle to train the target image on.

Expand source code

class Evolution:
    """Logic for a Species Evolution.

    Use the Evolution class when you want to train a Specie to look like
    a target image.

    Attributes:
        size (tuple): tuple containing height and width of target image (h, w).
        target (np.ndarray): target image for evolution.
        genes (int): the amount of circle to train the target image on.
        generation (int): amount of generations Evolution class has trained.
        specie (species.Specie): the Specie that is getting trained.
    """

    def __init__(self, size, target, genes=5):
        """Initializes Evolution class.

        Args:
            size (tuple): tuple containing height and width of target image (h, w).
            target (np.ndarray): target image for evolution.
            genes (int): the amount of circle to train the target image on.
        """
        self.size = size  # Tuple (y, x)
        self.target = target  # Target Image
        self.generation = 1
        self.genes = genes

        self.specie = Specie(size=self.size, genes=genes)

    def mutate(self, specie):
        """Mutates specie for evolution.

        Args:
            specie (species.Specie): Specie to mutate.

        Returns:
            New Specie class, that has been mutated.
        """
        new_specie = Specie(size=self.size, genotype=np.array(specie.genotype))

        # Randomization for Evolution
        y = random.randint(0, self.genes - 1)
        change = random.randint(0, 6)

        if change >= 6:
            change -= 1
            i, j = y, random.randint(0, self.genes - 1)
            i, j, s = (i, j, -1) if i < j else (j, i, 1)
            new_specie.genotype[i : j + 1] = np.roll(new_specie.genotype[i : j + 1], shift=s, axis=0)
            y = j

        selection = np.random.choice(5, size=change, replace=False)

        if random.random() < 0.25:
            new_specie.genotype[y, selection] = np.random.rand(len(selection))
        else:
            new_specie.genotype[y, selection] += (np.random.rand(len(selection)) - 0.5) / 3
            new_specie.genotype[y, selection] = np.clip(new_specie.genotype[y, selection], 0, 1)

        return new_specie

    def print_progress(self, fit):
        """Prints progress of Evolution.

        Args:
            fit (float): fitness value of specie.
        """
        print("GEN {}, FIT {:.8f}".format(self.generation, fit))

    def evolve(self, fitness=fitness.MSEFitness, max_generation=100000):
        """Genetic Algorithm for evolution.

        Call this function to begin evolving a Specie.

        Args:
            fitness (fitness.Fitness): fitness class to score species preformance.
            max_generation (int): amount of generations to train for.
        """
        fitness = fitness(self.target)

        self.specie.render()
        fit = fitness.score(self.specie.phenotype)

        for i in range(max_generation):
            self.generation = i + 1

            mutated = self.mutate(self.specie)
            mutated.render()
            newfit = fitness.score(mutated.phenotype)

            if newfit > fit:
                fit = newfit
                self.specie = mutated
                self.print_progress(newfit)

Methods

def evolve(self, fitness=circle_evolution.fitness.MSEFitness, max_generation=100000)

Genetic Algorithm for evolution.

Call this function to begin evolving a Specie.

Args

fitness : fitness.Fitness: fitness class to score species preformance.
max_generation : int: amount of generations to train for.

Expand source code

def evolve(self, fitness=fitness.MSEFitness, max_generation=100000):
    """Genetic Algorithm for evolution.

    Call this function to begin evolving a Specie.

    Args:
        fitness (fitness.Fitness): fitness class to score species preformance.
        max_generation (int): amount of generations to train for.
    """
    fitness = fitness(self.target)

    self.specie.render()
    fit = fitness.score(self.specie.phenotype)

    for i in range(max_generation):
        self.generation = i + 1

        mutated = self.mutate(self.specie)
        mutated.render()
        newfit = fitness.score(mutated.phenotype)

        if newfit > fit:
            fit = newfit
            self.specie = mutated
            self.print_progress(newfit)

def mutate(self, specie)

Mutates specie for evolution.

Args

specie : species.Specie: Specie to mutate.

Returns

New Specie class, that has been mutated.

Expand source code

def mutate(self, specie):
    """Mutates specie for evolution.

    Args:
        specie (species.Specie): Specie to mutate.

    Returns:
        New Specie class, that has been mutated.
    """
    new_specie = Specie(size=self.size, genotype=np.array(specie.genotype))

    # Randomization for Evolution
    y = random.randint(0, self.genes - 1)
    change = random.randint(0, 6)

    if change >= 6:
        change -= 1
        i, j = y, random.randint(0, self.genes - 1)
        i, j, s = (i, j, -1) if i < j else (j, i, 1)
        new_specie.genotype[i : j + 1] = np.roll(new_specie.genotype[i : j + 1], shift=s, axis=0)
        y = j

    selection = np.random.choice(5, size=change, replace=False)

    if random.random() < 0.25:
        new_specie.genotype[y, selection] = np.random.rand(len(selection))
    else:
        new_specie.genotype[y, selection] += (np.random.rand(len(selection)) - 0.5) / 3
        new_specie.genotype[y, selection] = np.clip(new_specie.genotype[y, selection], 0, 1)

    return new_specie

def print_progress(self, fit)

Prints progress of Evolution.

Args

fit : float: fitness value of specie.

Expand source code

def print_progress(self, fit):
    """Prints progress of Evolution.

    Args:
        fit (float): fitness value of specie.
    """
    print("GEN {}, FIT {:.8f}".format(self.generation, fit))