遺伝的アルゴリズムの基礎１

import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import sys
from IPython.display import clear_output

def create_choromosome(GENES_LENGTH):
    """
    染色体クラス。0か1の情報遺伝子をGENES_LENGTH分持つ。

    input:
        GENES_LENGTH(int): １染色体に入る遺伝子の数

    output:
        choromosome(np.array): 染色体

    note:
        今回の遺伝子情報は0と1だけなので遺伝子を計算する関数は省く。
    """
    return np.random.randint(0, 2, GENES_LENGTH)

def create_individual(GENES_LENGTH):
    """
    個体を作成する関数。今回の個体は染色体がひとつなので、染色体＝個体になっている
    """
    choromosome = create_choromosome(GENES_LENGTH)
    
    return choromosome

def create_generation(POPURATIONS_NUMBER, GENES_LENGTH):
    """
    世代を作る関数。

    input:
        POPURATIONS_NUMBER: 1世代の個体数
        GENES_LENGTH(int): 遺伝子の長さ(n)
    
    output: 
        generation(list): 個体クラスのリスト
    """
    generation = []
    for i in range(POPURATIONS_NUMBER):
        individual = create_individual(GENES_LENGTH)
        generation.append(individual)

    return generation

def calc_fitness(choromosome):
    """ 適応度を計算する。"""
    fitness = choromosome.sum() / len(choromosome)
    return fitness

def select_roulette(generation):
    '''
    ルーレット方式の親選択

    input:
        generation(list)
    
    output:
        parents(np.array)
    '''
    weights = [calc_fitness(individual) for individual in generation]
    norm_weights = [fitness / sum(weights) for fitness in weights]
    
    idxs = [i for i in range(len(generation))]
    idx = np.random.choice(idxs, p=norm_weights)
    
    parent = generation[idx]
    return parent
    
def select_tournament(generation, TOURNAMENT_SIZE):
    '''
    トーナメント方式の親選択

    input:
        generation(list)
        TOURNAMENT_SIZE(int): ランダムに抽選される人数。
        size(int): outputする親の数。デフォルトでは2人
    
    output:
        oarrents([np.array,...])
    '''
    idxs = [i for i in range(len(generation))]
    random_idxs = np.random.choice(idxs, TOURNAMENT_SIZE, replace=False)
    random_parents = [generation[idx] for idx in random_idxs]
    parrent = max(random_parents, key=calc_fitness)
    
    return parrent

def crossover_two_point(mother,father):
    '''
    二点交叉をして子孫を返す。
    
    入力:
        混ぜ合わせたい個体のペア
    出力:
        交叉後の個体のペア
    '''
    size = len(mother)
    point_befor = np.random.randint(0, size)
    point_after = np.random.randint(0, size - 1)

    if point_after >= point_befor:
        point_after += 1
    else:
        point_befor, point_after = point_after, point_befor

    choromosome = father
    choromosome[point_befor:point_after] = mother[point_befor:point_after]
    child = choromosome

    return child

def crossover_uniform(mother,father):
    '''
    一様交叉をして子孫を返す。
    
    入力:
        混ぜ合わせたい個体のペア
    出力:
        交叉後の個体のペア
    '''
    choromosome = []
    for i in range(len(mother)):
        if np.random.random() < 0.5:
            choromosome.append(father[i])
        else:
            choromosome.append(mother[i])

    child = np.array(choromosome)

    return child

def mutation(child,MUTATION_PLOB):
    if np.random.rand() < MUTATION_PLOB*0.01:
        i = np.random.randint(0, len(child)-1)
        child[i] = int(not child[i])
    else:
        return child

    return child

# param
POPURATIONS_NUMBER = 100 #@param {type:"number"}
GENES_LENGTH = 100 #@param {type:"number"}
GENERATION_COUNT = 20 #@param {type:"number"}
TOURNAMENT_SIZE = 3 #@param {type:"number"}
MUTATION_PLOB = 1 #@param {type:"number"}


# 祖先となる世代を作成
generation = create_generation(POPURATIONS_NUMBER, GENES_LENGTH)
plot_max = []
plot_min = []

# ここから進化開始
for generation_number in range(GENERATION_COUNT):
    next_generation = []

    for i in range(len(generation)):
        # mother = select_roulette(generation)
        # father = select_roulette(generation)
        mother = select_tournament(generation, TOURNAMENT_SIZE)
        father = select_tournament(generation, TOURNAMENT_SIZE)
        
        # child = crossover_two_point(mother,father)
        child = crossover_uniform(mother,father)

        child = mutation(child,MUTATION_PLOB)

        next_generation.append(child)
    
    fitness = [calc_fitness(choromosome) for choromosome in generation]
    clear_output(wait=True)
    print("第",generation_number+1,"/",GENERATION_COUNT,"世代")
    print("適応度 : 最大",max(fitness),", 最小",min(fitness))
    plot_max.append(max(fitness))
    plot_min.append(min(fitness))

    generation = next_generation

plt.plot(plot_max)
plt.plot(plot_min)
plt.show()

第 20 / 20 世代 適応度 : 最大 1.0 , 最小 0.97

<Figure size 432x288 with 1 Axes>