Python遺傳算法框架DEAP-Operators and Algorithms

Before starting with complex algorithms, we will see some basics of DEAP. First, we will start by creating simple individuals (as seen in the Creating Types tutorial) and make them interact with each other using different operators. Afterwards, we will learn how to use the algorithms and other tools.
在開始複雜的算法以前，咱們將會看到一些基礎的東西。首先，咱們開始建立簡單的個體（以前的教程中介紹了）和讓它們利用操做算子互相做用。而後，咱們將學習怎麼使用算法和其餘工具。

A First Individual（第一個個體）

First import the required modules and register the different functions required to create individuals that are lists of floats with a minimizing two objectives fitness.
一開始導入須要模塊和註冊不一樣的函數去建立個體（浮點數列表），求解的是一個最小化二目標適應度問題。

import random
 
from deap import base
from deap import creator
from deap import tools
 
IND_SIZE = 5
 
creator.create("FitnessMin", base.Fitness, weights=(-1.0, -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, n=IND_SIZE)

The first individual can now be built by adding the appropriate line to the script.
第一個個體如今能用合適一行代碼去建立添加了。

ind1 = toolbox.individual()

Printing the individual ind1 and checking if its fitness is valid will give something like this
打印出個體ind1,檢查它的適應度是否有效：

print ind1               # [0.86..., 0.27..., 0.70..., 0.03..., 0.87...]
print ind1.fitness.valid # False

The individual is printed as its base class representation (here a list) and the fitness is invalid because it contains no values.
這個個體打印出來了。這個適應度是無效的，由於沒有值。

Evaluation（重頭戲來了，評價函數/適應度函數）

The evaluation is the most personal part of an evolutionary algorithm, it is the only part of the library that you must write yourself. A typical evaluation function takes one individual as argument and returns its fitness as a tuple. As shown in the in the core section, a fitness is a list of floating point values and has a property valid to know if this individual shall be re-evaluated. The fitness is set by setting the values to the associated tuple. For example, the following evaluates the previously created individual ind1 and assigns its fitness to the corresponding values.
評價函數部分是評價算法中最「我的」的部分，它是庫中你必須本身寫的一部分。一個典型的評價函數須要一個個體做爲參數，而後返回這個個體的fitness（以tuple格式）。在覈心部分所示,一個適應度值是一個浮點值的列表並有一個有效的屬性來知道這個個體是否應當從新評估。這個適應度值是經過設置值和元祖關聯。舉個例子，下面的選擇評價以前建立的個體ind1，同時，分配它的fitness給相應的值。
(⊙o⊙)…讓我回憶起專業英語翻譯文章的感受了，難以讀懂。硬着頭皮往下看吧。

def evaluate(individual):
    # Do some hard computing on the individual
    a = sum(individual)
    b = len(individual)
    return a, 1. / b
 
ind1.fitness.values = evaluate(ind1)
print ind1.fitness.valid    # True    原來有`fitness`是True!搜嘎
print ind1.fitness          # (2.73, 0.2)上面建立的個體IND_SIZE=5，因此這裏是1/5=0.2

Dealing with single objective fitness is not different, the evaluation function must return a tuple because single-objective is treated as a special case of multi-objective.
與處理單目標適應度不一樣，這裏評價函數必須返回一個元祖由於單目標是做爲多目標的一個特例而已。

Mutation（變異）

There is a variety of mutation operators in the deap.tools module. Each mutation has its own characteristics and may be applied to different types of individuals. Be careful to read the documentation of the selected operator in order to avoid undesirable behaviour.
在deap.tools module模塊那兒有不少變異操做算子。每一個變異算子有它本身的特徵，能夠應用到不一樣類型的個體上。仔細閱讀選擇算子的文檔，從而避免沒法理解的行爲。

The general rule for mutation operators is that they only mutate, this means that an independent copy must be made prior to mutating the individual if the original individual has to be kept or is a reference to another individual (see the selection operator).
變異算子的整體規則就是他們只是變異，這意味着一個獨立的副本必須在變異個體操做以前，若是原始個體必需要保存或者是另外個體的參考（看選擇算子）。

In order to apply a mutation (here a gaussian mutation) on the individual ind1, simply apply the desired function.
爲了在個體ind1上施行變異（這裏是高斯變異），簡單應用了所需的函數。

mutant = toolbox.clone(ind1)
ind2, = tools.mutGaussian(mutant, mu=0.0, sigma=0.2, indpb=0.2)
del mutant.fitness.values

The fitness’ values are deleted because they’re not related to the individual anymore. As stated above, the mutation does mutate and only mutate an individual it is neither responsible of invalidating the fitness nor anything else. The following shows that ind2 and mutant are in fact the same individual.
適應度值被刪除了，由於它們再也不和這個個體相關了（由於變異了嘛！！！）。如上面所述，這個變異算子只是變異而且只是變異一個個體，它也不對適應度的無效負責或者其它。下面展現了ind2，變異算子事實上一樣的個體。（不是很明白這裏）

print ind2 is mutant    # True
print mutant is ind1    # False

Crossover（交叉）

The second kind of operator that we will present is the crossover operator. There is a variety of crossover operators in the deap.tools module. Each crossover has its own characteristics and may be applied to different types of individuals. Be careful to read the documentation of the selected operator in order to avoid undesirable behaviour.
咱們將要展現的第二種操做算子是交叉算子。在deap.tools module模塊中有不少交叉算子。每種交叉算子有它本身的特性，可能適用於不一樣類型的個體。仔細閱讀選擇算子的文檔從而避免沒法理解的的行爲。

The general rule for crossover operators is that they only mate individuals, this means that an independent copies must be made prior to mating the individuals if the original individuals have to be kept or are references to other individuals (see the selection operator).
交叉算子的整體規則是它們值負責交叉個體，這意味着在交叉個體以前需獨立的副本，若是原始個體不得不被保存或者做爲其餘個體的參考（看選擇算子）。

Lets apply a crossover operation to produce the two children that are cloned beforehand.
讓咱們施行交叉算子來產生兩個子孫（事先已經被「克隆」了）。

child1, child2 = [toolbox.clone(ind) for ind in (ind1, ind2)]
tools.cxBlend(child1, child2, 0.5)
del child1.fitness.values
del child2.fitness.values

Selection(選擇）

Selection is made among a population by the selection operators that are available in the deap.tools module. The selection operator usually takes as first argument an iterable container of individuals and the number of individuals to select. It returns a list containing the references to the selected individuals. The selection is made as follow.
選擇是經過選擇算子在種羣中操做的，它能夠在deap.tools module模塊中得到。這個選擇算子操做一般第一個參數是個體們的可迭代容器，以及要選擇個體的個數。它返回的是選擇個體們的應用列表。選擇算子操做示例以下：

selected = tools.selBest([child1, child2], 2)
print child1 in selected    # True

Usually duplication of the entire population will be made after selection or before variation.
一般在選擇或者變異以後，會作整個種羣的副本。

selected = toolbox.select(population, LAMBDA)
offspring = [toolbox.clone(ind) for ind in selected]

Using the Toolbox（使用工具箱）

The toolbox is intended to contain all the evolutionary tools, from the object initializers to the evaluation operator. It allows easy configuration of each algorithm. The toolbox has basically two methods, register() and unregister(), that are used to add or remove tools from the toolbox.
工具箱是來包含全部進化工具的，從對象的初始化到評價操做算子。它容許每種算法的配置。工具箱有兩個基本的方法，register()和unregister()，這個用來從工具箱中增長或者移除工具的。

This part of the tutorial will focus on registration of the evolutionary tools in the toolbox rather than the initialization tools. The usual names for the evolutionary tools are mate(), mutate(), evaluate() and select(), however, any name can be registered as long as it is unique. Here is how they are registered in the toolbox.
教程的這部分將會專一於工具箱中進化工具的註冊而不是初始化工具。進化工具一般的名字是mate(),mutate(),evaluate(),select()，然而，任何名字都是能夠被註冊的，只要它是獨一無二的。這裏展現了它們是怎麼被註冊進工具箱的：

from deap import base
from deap import tools

toolbox = base.Toolbox()

def evaluateInd(individual):
    # Do some computation
    return result,

toolbox.register("mate", tools.cxTwoPoint)
toolbox.register("mutate", tools.mutGaussian, mu=0, sigma=1, indpb=0.2)
toolbox.register("select", tools.selTournament, tournsize=3)
toolbox.register("evaluate", evaluateInd)

Using the toolbox for registering tools helps keeping the rest of the algorithms independent from the operator set. Using this scheme makes it very easy to locate and change any tool in the toolbox if needed.
使用工具箱註冊工具備助於保持其他算法獨立於操做符的集合。使用這種方案使得它很容易定位和改變工具箱中的任何工具。

Using the Tools（使用工具）

When building evolutionary algorithms the toolbox is used to contain the operators, which are called using their generic name. For example, here is a very simple generational evolutionary algorithm.
當創建進化算法時，工具箱是用來容納這個操做算子的，就是被稱爲使用它們的通用名字。例如，這裏是一個很是簡單的進化算法。

for g in range(NGEN):
    # Select the next generation individuals
    offspring = toolbox.select(pop, len(pop))
    # Clone the selected individuals
    offspring = map(toolbox.clone, offspring)

    # Apply crossover on the offspring
    for child1, child2 in zip(offspring[::2], offspring[1::2]):
        if random.random() < CXPB:
            toolbox.mate(child1, child2)
            del child1.fitness.values
            del child2.fitness.values

    # Apply mutation on the offspring
    for mutant in offspring:
        if random.random() < MUTPB:
            toolbox.mutate(mutant)
            del mutant.fitness.values

    # Evaluate the individuals with an invalid fitness
    invalid_ind = [ind for ind in offspring if not ind.fitness.valid]
    fitnesses = toolbox.map(toolbox.evaluate, invalid_ind)
    for ind, fit in zip(invalid_ind, fitnesses):
        ind.fitness.values = fit

    # The population is entirely replaced by the offspring
    pop[:] = offspring

This is a complete algorithm. It is generic enough to accept any kind of individual and any operator, as long as the operators are suitable for the chosen individual type. As shown in the last example, the usage of the toolbox allows to write algorithms that are as close as possible to pseudo code. Now it is up to you to write and experiment on your own.
這是一個完整的算法。它是足夠通用的來接收各類個體和任何操做算子，只要操做算子是是適合選擇的個體類型。正如最後一個例子所示，工具箱的使用容許寫算法（和僞代碼儘量相近）。如今取決於你本身寫、實驗你的算法。

Tool Decoration（工具裝飾）

Tool decoration is a very powerful feature that helps to control very precise things during an evolution without changing anything in the algorithm or operators. A decorator is a wrapper that is called instead of a function. It is asked to make some initialization and termination work before and after the actual function is called.
工具裝飾是一個很是強大的特性，在進化過程當中它有助於控制很是精確的東西而不須要改變算法或者算子的其餘東西。一個裝飾器被稱爲一個包裝器而不是一個函數。它是在一些初始化和終止工做的時候被訪問，在實際函數調用以前或者以後。

For example, in the case of a constrained domain, one can apply a decorator to the mutation and crossover in order to keep any individual from being out-of-bound. The following defines a decorator that checks if any attribute in the list is out-of-bound and clips it if this is the case. The decorator is defined using three functions in order to receive the min and max arguments. Whenever the mutation or crossover is called, bounds will be checked on the resulting individuals.
例如，在約束域的例子裏，一個能對變異和交叉引用裝飾器爲了不個體出界，若是出界就去除它。裝飾器使用三個函數來定義爲了接受最小和最大的參數。變異或交叉操做不管何時被調用，都會對結果個體們進行範圍檢查。
╮(╯▽╰)╭無奈，不明白啊，加油

def checkBounds(min, max):
    def decorator(func):
        def wrapper(*args, **kargs):
            offspring = func(*args, **kargs)
            for child in offspring:
                for i in xrange(len(child)):
                    if child[i] > max:
                        child[i] = max
                    elif child[i] < min:
                        child[i] = min
            return offspring
        return wrapper
    return decorator

toolbox.register("mate", tools.cxBlend, alpha=0.2)
toolbox.register("mutate", tools.mutGaussian, mu=0, sigma=2)

toolbox.decorate("mate", checkBounds(MIN, MAX))
toolbox.decorate("mutate", checkBounds(MIN, MAX))

This will work on crossover and mutation because both return a tuple of individuals. The mutation is often considered to return a single individual but again like for the evaluation, the single individual case is a special case of the multiple individual case.這將會做用在交叉和變異算子上由於二者都返回一個個體們的元祖。變異一般被認爲返回一個單一個體可是卻像評估算子同樣，單一個體的例子是多目標的一個特殊例子。