Python 遗传算法实现字符串

流程

1. 初始化

2. 适应度函数

3. 选择

4. 交叉

5. 变异

适应度函数计算方法

计算个体间的差：分别计算每个元素与目标元素的差取平方和

种群：计算总体均值

效果图

小结

• 一开始我设计的选择方法是把低于平均的个体淘汰，但是这样操作会导致陷入局部最优，循环500次依旧没有结果，很难找到最优个体。后面仔细看书，用书上的锦标赛算法，提高了随机性，可以找到了目标序列。

• 根据流程，我发现经过几轮的遗传，种群规模会迅速下降，因为中间补充的个体数量无法抵消淘汰的个体数量。于是我根据种群规模设计了阶梯式的概率，以便可以维持种群规模。

• 对于上面的问题，我开始想在选择后立即补充
  即 y=100/7+0.05x（x为选择后剩余的种群数量），但是补充的内容无法实现最后的功能：1.补充随机新个体，这就和选择前的操作无异，种群适应度没有太大变化。2.补充选择后的个体，会导致陷入局部最优，种群发展速度慢。3.综合 1 2 补充，
  依旧没有明显效果。

• 书上的最优概率，（主要是突变的概率非常低）对于产生新个体的速度较慢，基因重组的效果感觉不够明显，很容易陷入局部最优和陷入一个死局面，需要许久才能跳出这个局面

• 后面我自己修改了概率，依旧有这种问题，目前还没解决。

• 这次的代码写的比较不合理，高耦合。有许多地方可以改进，避免大量重复循环，以提高程序执行效率

补充

• 关于补充选择淘汰的个体，我发现可以在cross，即基因重组中改变他后代产生的children个数，直到恢复到设置的种群规模。在基因突变部分，不再是往种群中添加新的个体，而是修改现有种群的基因。

#coding=utf-8
import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
import randombest = [72, 101, 108, 108, 111, 44, 105,99,106,98,64,102,111,120,109,97,105,108]
#最佳序列def get_ran(b=33,e=126):"""获取随机字符值，默认从可显示字符开始（33-126）"""return random.randint(b,e)def get_pos(cou=17):"""获取随机位置，默认到17"""return random.randint(0,cou)class Chromosome:"""染色体类"""def __init__(self,dna=None):"""初始化，可指定DNA序列顺序，默认DNA为随机序列。"""if dna is not None:self.dna = dna[:]else :self.dna = []for i in range(18):self.dna.append(get_ran())def to_asc(self):"""将DNA序列由int转为str"""str = ''for chrx in self.dna:str = str + chr(chrx)return strdef __str__(self):"""将DNA序列由int转为str"""str = ''for chrx in self.dna:str = str + chr(chrx)return strdef mutation(self,which=-1):"""变异操作，默认产生随机个变异位点，突变为随机值，可指定突变位置"""if which == -1 :i = get_pos()for x in range(i):self.dna[get_pos()] = get_ran()else :self.dna[which]=get_pos()def comp(self,other):"""计算与指定序列DNA的平方差"""l = []val = 0for x in range(18):#print(x)d = self.dna[x] - other.dna[x]d = pow(d,2)l.append(d)val = val + dreturn l,valdef cross(self,other):"""该染色体与其他染色体基因重组，产生新染色体，采用多点交叉"""new = []for i in range(18):if get_pos(1) == 0:new.append(self.dna[i])else :new.append(other.dna[i])return Chromosome(new)class Population:"""种群类"""def __init__(self,much=5000,aim=[72, 101, 108, 108, 111, 44, 105,99,106,98,64,102,111,120,109,97,105,108]):"""初始化种群，默认种群数量5000，指定序列为 Hello,icjb@foxmail"""self.group = []self.best = Chromosome(aim)i = 0while i < much:self.group.append(Chromosome())i = i + 1def choice(self,p=0.05):"""选择操作，采用锦标赛选择方法，默认保留0.05"""group = []old = self.group[:]count = int(self.group.__len__() * p)once = int(self.group.__len__() / count)t_b = old[0]t = 0for ch in old:if t == once:group.append(t_b)t_b = cht = 0_,v1 = t_b.comp(self.best)_,v2 = ch.comp(self.best)if v1 >= v2 :t_b = cht = t + 1self.group.clear()self.group = group[:]def cross(self,p=0.7):"""交叉操作，采用多点交叉"""count = int(self.group.__len__()*p)i = 0group = []while i < count:t = self.group.__len__()group.append(self.group[get_pos(t-1)].cross(self.group[get_pos(t-1)]))i = i + 1self.group = self.group + groupdef mutation(self,p=0.001):"""种群突变"""count = int(self.group.__len__()*p)i = 0t = self.group.__len__()while i < count:self.group[get_pos(t-1)].mutation()i = i + 1def have_the_best(self):"""判断是否存在目标序列"""for ch in self.group:_,v = ch.comp(self.best)if v == 0 :return Truereturn Falsedef value(self):"""计算该种群的估值"""val = 0for ch in self.group:_,v = ch.comp(self.best)val = val + vreturn val / self.group.__len__()def get_best(self):"""获取种群最优个体"""best_one = self.group[0]_,val = best_one.comp(self.best)for ch in self.group:_,v = ch.comp(self.best)if val >= v :best_one = chval = vreturn best_onedef big_up(self):self.choice(0.9)self.cross(1.5)self.mutation(0.01)def big_down(self):self.choice()self.cross()self.mutation()def main():"""主函数"""ch_b = Chromosome(best)po = Population(5000)i = 1vals = []popu = []while i <= 500:val = po.value()leng = po.group.__len__()str1 = '第 {:^4} 代: 数量为 {:^6},  估值 :{:^6}'.format(i,leng,int(val))str2 = ' -- 最佳 : {}'.format(po.get_best().to_asc())if po.group.__len__() < 500:#种群个体少于500，需要补充大量新个体以免灭亡p1 = 0.8p2 = 1p3 = 0.12elif po.group.__len__() < 2500:p1 = 0.6p2 = 0.99p3 = 0.1elif po.group.__len__() < 5000:p1 = 0.5p2 = 0.9p3 = 0.09elif po.group.__len__() < 8000:p1 = 0.3p2 = 0.82p3 = 0.08elif po.group.__len__() < 10000:p1 = 0.2p2 = 0.78p3 = 0.05else :#默认操作p1 = 0.05p2 = 0.7p3 = 0.001# #但估值趋于稳定时，刺激种群# if val < 1000 :#     p1 = p1 - 0.5#     if p1 <= 0 :#         p1 = 0.05#     p2 = p2 + 0.3#     p3 = p3 + 0.05#模拟突增突降，可选# if get_pos(100) < 5 :#     po.big_down()#     print('big-down')# if get_pos(100) > 95 :#     po.big_up()#     print('big-up')print(str1+str2)#print(str2)po.choice(p1)if po.have_the_best() :print('找到目标！')print(ch_b.to_asc())breakpo.cross(p2)po.mutation(p3)vals.append(val)popu.append(leng)i = i + 1#绘图plt.figure(1)plt.subplot(121)x = range(i-1)y = vals[:]plt.plot(x,y)plt.subplot(122)y = popu[:]plt.plot(x,y)plt.show()if __name__ == "__main__":""""""main()input()

参考

遗传算法（python版） - 远方不远 - CSDN博客

一文读懂遗传算法工作原理（附Python实现）

【算法】超详细的遗传算法(Genetic Algorithm)解析 - 简书

遗传算法学习笔记（一）：常用的选择策略 - 依然传奇 - 博客园

遗传算法选择策略比较 - 道客巴巴

基于遗传算法的人工智能实例之拼图游戏（python实现） - 不基调的博客 - CSDN博客

另一版本实现方式，改自GITHUB：

import numpy as np
import matplotlib as mpl
import matplotlib.pyplot as plt
import timefrom random import (choice, random, randint)__all__ = ['Chromosome', 'Population']class Chromosome:_target_gene = "Hello, icjb@foxmail.com"def __init__(self, gene):self.gene = geneself.fitness = Chromosome._update_fitness(gene)def mate(self, mate):pivot = randint(0, len(self.gene) - 1)gene1 = self.gene[:pivot] + mate.gene[pivot:]gene2 = mate.gene[:pivot] + self.gene[pivot:]        return Chromosome(gene1), Chromosome(gene2)def mutate(self):gene = list(self.gene)delta = randint(44, 122)idx = randint(0, len(gene) - 1)gene[idx] = chr((ord(gene[idx]) + delta) % 123)return Chromosome(''.join(gene))@staticmethod            def _update_fitness(gene):fitness = 0for a, b in zip(gene, Chromosome._target_gene):fitness += abs(ord(a) - ord(b))          return fitness@staticmethoddef gen_random():gene = []for x in range(len(Chromosome._target_gene)):gene.append(chr(randint(44, 122)))return Chromosome(''.join(gene))class Population:_tournamentSize = 3def __init__(self, size=1024, crossover=0.8, elitism=0.1, mutation=0.3):self.elitism = elitismself.mutation = mutationself.crossover = crossoverbuf = []for i in range(size): buf.append(Chromosome.gen_random())self.population = list(sorted(buf, key=lambda x: x.fitness))def _tournament_selection(self):best = choice(self.population)for i in range(Population._tournamentSize):cont = choice(self.population)if (cont.fitness < best.fitness): best = contreturn bestdef _selectParents(self):                 return (self._tournament_selection(), self._tournament_selection())def evolve(self):size = len(self.population)idx = int(round(size * self.elitism))buf = self.population[:idx]while (idx < size):if random() <= self.crossover:(p1, p2) = self._selectParents()children = p1.mate(p2)for c in children:if random() <= self.mutation:buf.append(c.mutate())else:buf.append(c)idx += 2else:if random() <= self.mutation:buf.append(self.population[idx].mutate())else:buf.append(self.population[idx])idx += 1self.population = list(sorted(buf[:size], key=lambda x: x.fitness))if __name__ == "__main__":maxGenerations = 2000t1 = time.time()pop = Population(size=2000, crossover=0.7, elitism=0.05, mutation=0.9)li = []x = []for i in range(1, maxGenerations + 1):print("Generation %d  Fitness:%d  Result:%s" % (i, Chromosome._update_fitness(pop.population[0].gene),pop.population[0].gene) )if pop.population[0].fitness == 0: breakelse:pop.evolve()li.append(Chromosome._update_fitness(pop.population[0].gene))xx = 0for p in range(pop.population.__len__()):xx += Chromosome._update_fitness(pop.population[p].gene)xx = xx / pop.population.__len__()x.append(xx)x.sort()print(x[0])print("Maximum generations reached without success.")t2 = time.time()t = int(t2 - t1)print(f'+ 共用时： {t} s')plt.figure(1)plt.subplot(111)x = range(i-1)y = li[:]plt.plot(x,y)plt.xlabel("generation",fontsize=12)plt.ylabel("fitness",fontsize=12)plt.show()