Python人工智能

遗传算法

遗传算法详细操作教程
本章详细讨论AI的遗传算法。

什么是遗传算法?

遗传算法(GA)是基于自然选择和遗传学概念的基于搜索的算法。遗传算法是称为进化计算的更大分支的一个子集。
GAs由John Holland及其在密歇根大学的学生和同事开发,最着名的是David E.Goldberg。自那以来,它一直在尝试各种优化问题并取得了很高的成功。
在GAs中,我们为给定问题提供了一系列可能的解决方案。这些解决方案然后经历重组和突变(如在自然遗传学中),产生新的儿童,并且该过程在各代重复。每个个体(或候选解决方案)都被分配一个适应值(基于其目标函数值),并且适合者个体被赋予更高的配偶并产生更适合个体的机会。这符合达尔文适者生存理论。
因此,它不断发展更好的个人或解决方案,直到达到停止标准。
遗传算法在本质上具有充分的随机性,但它们比随机局部搜索(我们只是尝试随机解决方案,追踪迄今为止最好的)的性能好得多,因为它们也在利用历史信息。

如何使用遗传算法优化问题?

优化是使设计,状况,资源和系统尽可能有效。 以下框图显示了优化过程 -
遗传算法优化

GA机制优化过程的阶段

以下是用于优化问题的GA机制的一系列步骤。
第1步 - 随机生成初始群体。 第2步 - 选择具有最佳适应值的初始解决方案。 第3步 - 使用变异和交叉算子重组选定的解决方案。 第4步 - 将后代插入群体。 第5步 - 现在,如果停止条件得到满足,则返回具有最佳适应值的解。 否则,请转到第2步。

安装必要的软件包

要在Python中使用遗传算法来解决这个问题,我们将使用一个称为DEAP的功能强大的GA包。 它是用于快速建立原型和测试思想的新型演化计算框架库。在命令提示符下使用以下命令来安装此软件包 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
pip install deap
如果您使用的是anaconda环境,则可以使用以下命令安装deap -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
conda install -c conda-forge deap

使用遗传算法实现解决方案

本节向您介绍使用遗传算法实现解决方案。
生成位模式
以下示例显示了如何根据One Max问题生成一个包含15个字符串的位串。
如下所示导入必要的软件包 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
import random
from deap import base, creator, tools
定义评估函数。 这是创建遗传算法的第一步。
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
def eval_func(individual):
   target_sum = 15
   return len(individual) - abs(sum(individual) - target_sum)
现在,使用正确的参数创建工具箱 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
def create_toolbox(num_bits):
   creator.create("FitnessMax", base.Fitness, weights=(1.0,))
   creator.create("Individual", list, fitness=creator.FitnessMax)
初始化工具箱
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
toolbox = base.Toolbox()
toolbox.register("attr_bool", random.randint, 0, 1)
toolbox.register("individual", tools.initRepeat, creator.Individual,
   toolbox.attr_bool, num_bits)
toolbox.register("population", tools.initRepeat, list, toolbox.individual)
注册计算操作符 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
toolbox.register("evaluate", eval_func)
现在,注册交叉运算符 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
toolbox.register("mate", tools.cxTwoPoint)
注册一个可变运算符 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
toolbox.register("mutate", tools.mutFlipBit, indpb = 0.05)
定义育种操作符 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
toolbox.register("select", tools.selTournament, tournsize = 3)
return toolbox
if __name__ == "__main__":
   num_bits = 45
   toolbox = create_toolbox(num_bits)
   random.seed(7)
   population = toolbox.population(n = 500)
   probab_crossing, probab_mutating = 0.5, 0.2
   num_generations = 10
   print('\nEvolution process starts')
评估整个人口 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
fitnesses = list(map(toolbox.evaluate, population))
for ind, fit in zip(population, fitnesses):
   ind.fitness.values = fit
print('\nEvaluated', len(population), 'individuals')
经过几代人的创建和迭代 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
for g in range(num_generations):
   print("\n- Generation", g)
选择下一代个人 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
offspring = toolbox.select(population, len(population))
现在,克隆选定的个人 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
offspring = list(map(toolbox.clone, offspring))
对后代应用交叉和变异 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
for child1, child2 in zip(offspring[::2], offspring[1::2]):
   if random.random() < probab_crossing:
   toolbox.mate(child1, child2)
删除孩子的适应值
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
del child1.fitness.values
del child2.fitness.values
现在,应用突变 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
for mutant in offspring:
   if random.random() < probab_mutating:
   toolbox.mutate(mutant)
   del mutant.fitness.values
评估与无效的健身个体 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
invalid_ind = [ind for ind in offspring if not ind.fitness.valid]
fitnesses = map(toolbox.evaluate, invalid_ind)
for ind, fit in zip(invalid_ind, fitnesses):
   ind.fitness.values = fit
print('Evaluated', len(invalid_ind), 'individuals')
现在,用下一代个体替代人口 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
population[:] = offspring
打印当代人的统计数据 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
fits = [ind.fitness.values[0] for ind in population]
length = len(population)
mean = sum(fits) / length
sum2 = sum(x*x for x in fits)
std = abs(sum2 / length - mean**2)**0.5
print('Min =', min(fits), ', Max =', max(fits))
print('Average =', round(mean, 2), ', Standard deviation =',
round(std, 2))
print("\n- Evolution ends")
打印最终输出 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
best_ind = tools.selBest(population, 1)[0]
   print('\nBest individual:\n', best_ind)
   print('\nNumber of ones:', sum(best_ind))
Following would be the output:
Evolution process starts
Evaluated 500 individuals
- Generation 0
Evaluated 295 individuals
Min = 32.0 , Max = 45.0
Average = 40.29 , Standard deviation = 2.61
- Generation 1
Evaluated 292 individuals
Min = 34.0 , Max = 45.0
Average = 42.35 , Standard deviation = 1.91
- Generation 2
Evaluated 277 individuals
Min = 37.0 , Max = 45.0
Average = 43.39 , Standard deviation = 1.46
… … … …
- Generation 9
Evaluated 299 individuals
Min = 40.0 , Max = 45.0
Average = 44.12 , Standard deviation = 1.11
- Evolution ends
Best individual:
[0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 1,
 1, 0, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0,
 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 1]
Number of ones: 15
符号回归问题
这是遗传编程中最著名的问题之一。 所有符号回归问题都使用任意数据分布,并尝试用符号公式来拟合最准确的数据。 通常,像RMSE(均方根误差)这样的度量用于度量个体的适应度。 这是一个经典的回归问题,这里我们使用方程:5x3-6x2 + 8x = 1。 我们需要按照上述示例中的所有步骤进行操作,但主要部分是创建基元集,因为它们是个人的构建基块,因此可以开始评估。 这里将使用经典的基元集。
以下Python代码详细解释了这一点 -
# Filename : example.py
# Copyright : 2020 By Lidihuo
# Author by : www.lidihuo.com
# Date : 2020-08-26
import operator
import math
import random
import numpy as np
from deap import algorithms, base, creator, tools, gp
def division_operator(numerator, denominator):
   if denominator == 0:
      return 1
   return numerator / denominator
def eval_func(individual, points):
   func = toolbox.compile(expr=individual)
   return math.fsum(mse) / len(points),
def create_toolbox():
   pset = gp.PrimitiveSet("MAIN", 1)
   pset.addPrimitive(operator.add, 2)
   pset.addPrimitive(operator.sub, 2)
   pset.addPrimitive(operator.mul, 2)
   pset.addPrimitive(division_operator, 2)
   pset.addPrimitive(operator.neg, 1)
   pset.addPrimitive(math.cos, 1)
   pset.addPrimitive(math.sin, 1)
   pset.addEphemeralConstant("rand101", lambda: random.randint(-1,1))
   pset.renameArguments(ARG0 = 'x')
   creator.create("FitnessMin", base.Fitness, weights = (-1.0,))
   creator.create("Individual",gp.PrimitiveTree,fitness=creator.FitnessMin)
   toolbox = base.Toolbox()
   toolbox.register("expr", gp.genHalfAndHalf, pset=pset, min_=1, max_=2)
   toolbox.expr)
   toolbox.register("population",tools.initRepeat,list, toolbox.individual)
   toolbox.register("compile", gp.compile, pset = pset)
   toolbox.register("evaluate", eval_func, points = [x/10. for x in range(-10,10)])
   toolbox.register("select", tools.selTournament, tournsize = 3)
   toolbox.register("mate", gp.cxOnePoint)
   toolbox.register("expr_mut", gp.genFull, min_=0, max_=2)
   toolbox.register("mutate", gp.mutUniform, expr = toolbox.expr_mut, pset = pset)
   toolbox.decorate("mate", gp.staticLimit(key = operator.attrgetter("height"), max_value = 17))
   toolbox.decorate("mutate", gp.staticLimit(key = operator.attrgetter("height"), max_value = 17))
   return toolbox
if __name__ == "__main__":
   random.seed(7)
   toolbox = create_toolbox()
   population = toolbox.population(n = 450)
   hall_of_fame = tools.HallOfFame(1)
   stats_fit = tools.Statistics(lambda x: x.fitness.values)
   stats_size = tools.Statistics(len)
   mstats = tools.MultiStatistics(fitness=stats_fit, size = stats_size)
   mstats.register("avg", np.mean)
   mstats.register("std", np.std)
   mstats.register("min", np.min)
   mstats.register("max", np.max)
   probab_crossover = 0.4
   probab_mutate = 0.2
   number_gen = 10
   population, log = algorithms.eaSimple(population, toolbox,
      probab_crossover, probab_mutate, number_gen,
      stats = mstats, halloffame = hall_of_fame, verbose = True)
请注意,所有基本步骤与生成位模式时使用的步骤相同。 这个程序会给出10代后的输出为min,max,std(标准偏差)。
昵称: 邮箱:
Copyright © 2022 立地货 All Rights Reserved.
备案号:京ICP备14037608号-4