【python(deap库)实现】GEAP 遗传算法/遗传编程 genetic programming +

巴壁虎

2020-05-17

前言

本文不介绍原理的东西，主要是实现进化算法的python实现。
原理介绍可以看这里，能学习要很多，我也在这里写了一些感受心得：
遗传算法/遗传编程进化算法基于python DEAP库深度解析讲解

1.优化问题的定义

单目标优化

creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0, ))

在创建单目标优化问题时，weights用来指示最大化和最小化。此处-1.0即代表问题是一个最小化问题，对于最大化，应将weights改为正数，如1.0。
另外即使是单目标优化，weights也需要是一个tuple，以保证单目标和多目标优化时数据结构的统一。
对于单目标优化问题，weights 的绝对值没有意义，只要符号选择正确即可。

多目标优化

creator.create(‘FitnessMulti‘, base.Fitness, weights=(-1.0, 1.0))

对于多目标优化问题，weights用来指示多个优化目标之间的相对重要程度以及最大化最小化。如示例中给出的(-1.0, 1.0)代表对第一个目标函数取最小值，对第二个目标函数取最大值。

2.个体编码

实数编码(Value encoding)：直接用实数对变量进行编码。优点是不用解码，基因表达非常简洁，而且能对应连续区间。但是实数编码后搜索区间连续，因此容易陷入局部最优。

实数编码

from deap import base, creator, tools
import random
IND_SIZE = 5
creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0,)) #优化目标：单变量，求最小值
creator.create(‘Individual‘, list, fitness = creator.FitnessMin) #创建Individual类，继承list

toolbox = base.Toolbox()
toolbox.register(‘Attr_float‘, random.random)
toolbox.register(‘Individual‘, tools.initRepeat, creator.Individual, toolbox.Attr_float, n=IND_SIZE)

ind1 = toolbox.Individual()
print(ind1)

# 结果：[0.8579615693371493, 0.05774821674048369, 0.8812411734389638, 0.5854279538236896, 0.12908399219828248]

二进制编码

from deap import base, creator, tools
from scipy.stats import bernoulli

creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0,)) #优化目标：单变量，求最小值
creator.create(‘Individual‘, list, fitness = creator.FitnessMin) #创建Individual类，继承list

GENE_LENGTH = 10

toolbox = base.Toolbox()
toolbox.register(‘Binary‘, bernoulli.rvs, 0.5) #注册一个Binary的alias，指向scipy.stats中的bernoulli.rvs，概率为0.5
toolbox.register(‘Individual‘, tools.initRepeat, creator.Individual, toolbox.Binary, n = GENE_LENGTH) #用tools.initRepeat生成长度为GENE_LENGTH的Individual

ind1 = toolbox.Individual()
print(ind1)

# 结果：[1, 0, 0, 0, 0, 1, 0, 1, 1, 0]

序列编码(Permutation encoding)

from deap import base, creator, tools
import random
creator.create("FitnessMin", base.Fitness, weights=(-1.0,))
creator.create("Individual", list, fitness=creator.FitnessMin)

IND_SIZE=10

toolbox = base.Toolbox()
toolbox.register("Indices", random.sample, range(IND_SIZE), IND_SIZE)
toolbox.register("Individual", tools.initIterate, creator.Individual,toolbox.Indices)
ind1 = toolbox.Individual()
print(ind1)

#结果：[0, 1, 5, 8, 2, 3, 6, 7, 9, 4]

粒子(Particles)

import random
from deap import base, creator, tools

creator.create("FitnessMax", base.Fitness, weights=(1.0, 1.0))
creator.create("Particle", list, fitness=creator.FitnessMax, speed=None,
               smin=None, smax=None, best=None)

# 自定义的粒子初始化函数
def initParticle(pcls, size, pmin, pmax, smin, smax):
    part = pcls(random.uniform(pmin, pmax) for _ in range(size))
    part.speed = [random.uniform(smin, smax) for _ in range(size)]
    part.smin = smin
    part.smax = smax
    return part

toolbox = base.Toolbox()
toolbox.register("Particle", initParticle, creator.Particle, size=2, pmin=-6, pmax=6, smin=-3, smax=3) #为自己编写的initParticle函数注册一个alias "Particle"，调用时生成一个2维粒子，放在容器creator.Particle中，粒子的位置落在（-6,6）中，速度限制为（-3，3）

ind1 = toolbox.Particle()
print(ind1)
print(ind1.speed)
print(ind1.smin, ind1.smax)

# 结果：[-2.176528549934324, -3.0796558214905]
#[-2.9943676285620104, -0.3222138308543414]
#-3 3

print(ind1.fitness.valid)

# 结果：False
# 因为当前还没有计算适应度函数，所以粒子的最优适应度值还是invalid

3 初始种群建立

一般族群

这是最常用的族群类型，族群中没有特别的顺序或者子族群。

from deap import base, creator, tools
from scipy.stats import bernoulli

# 定义问题
creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0,)) # 单目标，最小化
creator.create(‘Individual‘, list, fitness = creator.FitnessMin)

# 生成个体
GENE_LENGTH = 5
toolbox = base.Toolbox() #实例化一个Toolbox
toolbox.register(‘Binary‘, bernoulli.rvs, 0.5)
toolbox.register(‘Individual‘, tools.initRepeat, creator.Individual, toolbox.Binary, n=GENE_LENGTH)

# 生成初始族群
N_POP = 10
toolbox.register(‘Population‘, tools.initRepeat, list, toolbox.Individual)
toolbox.Population(n = N_POP)

# 结果：
# [[1, 0, 1, 1, 0],
# [0, 1, 1, 0, 0],
# [0, 1, 0, 0, 0],
# [1, 1, 0, 1, 0],
# [0, 1, 1, 1, 1],
# [0, 1, 1, 1, 1],
# [1, 0, 0, 0, 1],
# [1, 1, 0, 1, 0],
# [0, 1, 1, 0, 1],
# [1, 0, 0, 0, 0]]

粒子群

粒子群中的所有粒子共享全局最优。在实现时需要额外传入全局最优位置与全局最优适应度给族群。

creator.create("Swarm", list, gbest=None, gbestfit=creator.FitnessMax)
toolbox.register("swarm", tools.initRepeat, creator.Swarm, toolbox.particle)

4 评价

评价部分是根据任务的特性高度定制的，DEAP库中并没有预置的评价函数模版。
在使用DEAP时，需要注意的是，无论是单目标还是多目标优化，评价函数的返回值必须是一个tuple类型。

from deap import base, creator, tools
import numpy as np
# 定义问题
creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0,)) #优化目标：单变量，求最小值
creator.create(‘Individual‘, list, fitness = creator.FitnessMin) #创建Individual类，继承list

# 生成个体
IND_SIZE = 5
toolbox = base.Toolbox()
toolbox.register(‘Attr_float‘, np.random.rand)
toolbox.register(‘Individual‘, tools.initRepeat, creator.Individual, toolbox.Attr_float, n=IND_SIZE)

# 生成初始族群
N_POP = 10
toolbox.register(‘Population‘, tools.initRepeat, list, toolbox.Individual)
pop = toolbox.Population(n = N_POP)

# 定义评价函数
def evaluate(individual):
  return sum(individual), #注意这个逗号，即使是单变量优化问题，也需要返回tuple

# 评价初始族群
toolbox.register(‘Evaluate‘, evaluate)
fitnesses = map(toolbox.Evaluate, pop)
for ind, fit in zip(pop, fitnesses):
  ind.fitness.values = fit
  print(ind.fitness.values)

# 结果：
# (2.593989197511478,)
# (1.1287944225903104,)
# (2.6030877077096717,)
# (3.304964061515382,)
# (2.534627558467466,)
# (2.4697149450205536,)
# (2.344837782191844,)
# (1.8959030773060852,)
# (2.5192475334239,)
# (3.5069764929866585,)

5 配种选择

selTournament() 锦标赛选择
selRoulette() 轮盘赌选择（不能用于最小化或者适应度会小于等于0的问题）
selNSGA2() NSGA-II选择，适用于多目标遗传算法
selSPEA2() SPEA2选择，目前版本(ver 1.2.2)的该函数实现有误，没有为个体分配距离，不建议使用。
selRandom() 有放回的随机选择
selBest() 选择最佳
selWorst() 选择最差
selTournamentDCD() Dominance/Crowding distance锦标赛选择，目前版本的实现也有些问题
selDoubleTournament() Size+Fitness双锦标赛选择
selStochasticUniversalSampling() 随机抽样选择
selLexicase() 词典选择，参考这篇文章
selEpsilonLexicase() 词典选择在连续值域上的扩展

from deap import base, creator, tools
import numpy as np
# 定义问题
creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0,)) #优化目标：单变量，求最小值
creator.create(‘Individual‘, list, fitness = creator.FitnessMin) #创建Individual类，继承list

# 生成个体
IND_SIZE = 5
toolbox = base.Toolbox()
toolbox.register(‘Attr_float‘, np.random.rand)
toolbox.register(‘Individual‘, tools.initRepeat, creator.Individual, toolbox.Attr_float, n=IND_SIZE)

# 生成初始族群
N_POP = 10
toolbox.register(‘Population‘, tools.initRepeat, list, toolbox.Individual)
pop = toolbox.Population(n = N_POP)

# 定义评价函数
def evaluate(individual):
  return sum(individual), #注意这个逗号，即使是单变量优化问题，也需要返回tuple

# 评价初始族群
toolbox.register(‘Evaluate‘, evaluate)
fitnesses = map(toolbox.Evaluate, pop)
for ind, fit in zip(pop, fitnesses):
  ind.fitness.values = fit

# 选择方式1：锦标赛选择
toolbox.register(‘TourSel‘, tools.selTournament, tournsize = 2) # 注册Tournsize为2的锦标赛选择
selectedTour = toolbox.TourSel(pop, 5) # 选择5个个体
print(‘锦标赛选择结果：‘)
for ind in selectedTour:
  print(ind)
  print(ind.fitness.values)

# 选择方式2: 轮盘赌选择
toolbox.register(‘RoulSel‘, tools.selRoulette)
selectedRoul = toolbox.RoulSel(pop, 5)
print(‘轮盘赌选择结果：‘)
for ind in selectedRoul:
  print(ind)
  print(ind.fitness.values)

# 选择方式3: 随机普遍抽样选择
toolbox.register(‘StoSel‘, tools.selStochasticUniversalSampling)
selectedSto = toolbox.StoSel(pop, 5)
print(‘随机普遍抽样选择结果：‘)
for ind in selectedSto:
  print(ind)
  print(ind.fitness.values)
  
#结果：
#锦标赛选择结果：
#[0.2673058115582905, 0.8131397980144155, 0.13627430737326807, 0.10792026110464248, 0.4166962522797264]
#(1.741336430330343,)
#[0.5448284697291571, 0.9702727117158071, 0.03349947770537576, 0.7018813286570782, 0.3244029157717422]
#(2.5748849035791603,)
#[0.8525836387058023, 0.28064482205939634, 0.9235436615033125, 0.6429467684175085, 0.5965523553349544]
#(3.296271246020974,)
#[0.5243293164960845, 0.37883291328325286, 0.28423194217619596, 0.5005947374376103, 0.3017896612109636]
#(1.9897785706041071,)
#[0.4038211036464676, 0.841374996509095, 0.3555644512425019, 0.5849111474726337, 0.058759891556433574]
#(2.2444315904271317,)
#轮盘赌选择结果：
#[0.42469039733882064, 0.8411201950346711, 0.6322812691061555, 0.7566549973076343, 0.9352307652371067]
#(3.5899776240243884,)
#[0.42469039733882064, 0.8411201950346711, 0.6322812691061555, 0.7566549973076343, 0.9352307652371067]
#(3.5899776240243884,)
#[0.5448284697291571, 0.9702727117158071, 0.03349947770537576, 0.7018813286570782, 0.3244029157717422]
#(2.5748849035791603,)
#[0.630305953330188, 0.09565983206218687, 0.890691659939096, 0.8706091807317707, 0.19708949882847437]
#(2.684356124891716,)
#[0.5961060867498598, 0.4300051776616509, 0.4512760237511251, 0.047731561819711055, 0.009892120639829804]
#(1.5350109706221766,)
#随机普遍抽样选择结果：
#[0.2673058115582905, 0.8131397980144155, 0.13627430737326807, 0.10792026110464248, 0.4166962522797264]
#(1.741336430330343,)
#[0.4038211036464676, 0.841374996509095, 0.3555644512425019, 0.5849111474726337, 0.058759891556433574]
#(2.2444315904271317,)
#[0.630305953330188, 0.09565983206218687, 0.890691659939096, 0.8706091807317707, 0.19708949882847437]
#(2.684356124891716,)
#[0.40659881466060876, 0.8387139101647804, 0.28504735705240236, 0.46171554118627334, 0.7843353275244066]
#(2.7764109505884718,)
#[0.42469039733882064, 0.8411201950346711, 0.6322812691061555, 0.7566549973076343, 0.9352307652371067]
#(3.5899776240243884,)

6 变异

cxOnePoint() 单点交叉实数、二进制
cxTwoPoint() 两点交叉实数、二进制
cxUniform() 均匀交叉实数、二进制
cxPartialyMatched() 部分匹配交叉PMX 序列
cxUniformPartialyMatched() PMX变种，改两点为均匀交叉序列
cxOrdered() 有序交叉序列
cxBlend() 混合交叉实数
cxESBlend() 带进化策略的混合交叉
cxESTwoPoint() 带进化策略的两点交叉
cxSimulatedBinary() 模拟二值交叉实数
cxSimulatedBinaryBounded() 有界模拟二值交叉实数
cxMessyOnePoint() 混乱单点交叉实数、二进制

from deap import base, creator, tools
import random
# 创建两个序列编码个体
random.seed(42) # 保证结果可复现
IND_SIZE = 8
creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0, ))
creator.create(‘Individual‘, list, fitness = creator.FitnessMin)

toolbox = base.Toolbox()
toolbox.register(‘Indices‘, random.sample, range(IND_SIZE), IND_SIZE)
toolbox.register(‘Individual‘, tools.initIterate, creator.Individual, toolbox.Indices)

ind1, ind2 = [toolbox.Individual() for _ in range(2)]
print(ind1, ‘\n‘, ind2)
# 结果：[1, 0, 5, 2, 7, 6, 4, 3] 
# [1, 4, 3, 0, 6, 5, 2, 7]

# 单点交叉
child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxOnePoint(child1, child2)
print(child1, ‘\n‘, child2)
#结果：[1, 4, 3, 0, 6, 5, 2, 7] 
# [1, 0, 5, 2, 7, 6, 4, 3]
# 可以看到从第四位开始被切开并交换了

# 两点交叉
child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxTwoPoint(child1, child2)
print(child1, ‘\n‘, child2)
# 结果：[1, 0, 5, 2, 6, 5, 2, 3] 
# [1, 4, 3, 0, 7, 6, 4, 7]
# 基因段[6, 5, 2]与[7, 6, 4]互换了

# 均匀交叉
child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxUniform(child1, child2, 0.5)
print(child1, ‘\n‘, child2)
# 结果：[1, 0, 3, 2, 7, 5, 4, 3] 
# [1, 4, 5, 0, 6, 6, 2, 7]

# 部分匹配交叉
child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxPartialyMatched(child1, child2)
print(child1, ‘\n‘, child2)
# 结果：[1, 0, 5, 2, 6, 7, 4, 3] 
# [1, 4, 3, 0, 7, 5, 2, 6]
# 可以看到与之前交叉算子的明显不同，这里的每个序列都没有冲突

# 有序交叉
child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxOrdered(child1, child2)
print(child1, ‘\n‘, child2)
# 结果：[5, 4, 3, 2, 7, 6, 1, 0] 
# [3, 0, 5, 6, 2, 7, 1, 4]

# 混乱单点交叉
child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxMessyOnePoint(child1, child2)
print(child1, ‘\n‘, child2)
# 结果：[1, 0, 5, 2, 7, 4, 3, 0, 6, 5, 2, 7] 
# [1, 6, 4, 3]
# 注意个体序列长度的改变

7 突变

from deap import base, creator, tools
import random
# 创建一个实数编码个体
random.seed(42) # 保证结果可复现
IND_SIZE = 5
creator.create(‘FitnessMin‘, base.Fitness, weights=(-1.0, ))
creator.create(‘Individual‘, list, fitness = creator.FitnessMin)

toolbox = base.Toolbox()
toolbox.register(‘Attr_float‘, random.random)
toolbox.register(‘Individual‘, tools.initRepeat, creator.Individual, toolbox.Attr_float, IND_SIZE)

ind1 = toolbox.Individual()
print(ind1)
# 结果：[0.6394267984578837, 0.025010755222666936, 0.27502931836911926, 0.22321073814882275, 0.7364712141640124]

# 高斯突变
mutant = toolbox.clone(ind1)
tools.mutGaussian(mutant, 3, 0.1, 1)
print(mutant)
# 结果：[3.672658632864655, 2.99827700737295, 3.2982590920597916, 3.339566606808737, 3.6626390539295306]
# 可以看到当均值给到3之后，变异形成的个体均值从0.5也增大到了3附近

# 乱序突变
mutant = toolbox.clone(ind1)
tools.mutShuffleIndexes(mutant, 0.5)
print(mutant)
# 结果：[0.22321073814882275, 0.7364712141640124, 0.025010755222666936, 0.6394267984578837, 0.27502931836911926]

# 有界多项式突变
mutant = toolbox.clone(ind1)
tools.mutPolynomialBounded(mutant, 20, 0, 1, 0.5)
print(mutant)
# 结果：[0.674443861742489, 0.020055418656044655, 0.2573977358171454, 0.11555018832942898, 0.6725269223692601]

# 均匀整数突变
mutant = toolbox.clone(ind1)
tools.mutUniformInt(mutant, 1, 5, 0.5)
print(mutant)
# 结果：[0.6394267984578837, 3, 0.27502931836911926, 0.22321073814882275, 0.7364712141640124]
# 可以看到在第二个位置生成了整数3