前往小程序,Get更优阅读体验!
立即前往
首页
学习
活动
专区
工具
TVP
发布
社区首页 >专栏 >代码 | 自适应大邻域搜索系列之(3) - Destroy和Repair方法代码实现解析

代码 | 自适应大邻域搜索系列之(3) - Destroy和Repair方法代码实现解析

作者头像
用户1621951
发布2019-10-18 02:05:56
5200
发布2019-10-18 02:05:56
举报
文章被收录于专栏:数据魔术师数据魔术师

写在前面

上一篇文章中我们具体解剖了ALNS类的具体代码实现过程,不过也留下了很多大坑。

接下来的文章基本都是“填坑”了,把各个模块一一展现解析给大家。

不过碍于文章篇幅等原因呢,也不会每一行代码都进行讲解,那些简单的代码就跳过了,相信大家也能一眼就看懂。好了,废话不多说,开始干活吧。

01 照旧总体概述

前面我们提到,ALNS中的重中之重就是Destroy和Repair方法了,在整一个ALNS框架中呢,涉及这俩货的有:

Destroy和Repair方法的具体实现、Destroy和Repair方法管理(包括各个Destroy和Repair方法权重分配、成绩打分、按权重选择哪个Destroy和Repair方法等操作)

所以在这次的ALNS代码中呢,这俩货的代码实现呢也分为两个模块:

  • Destroy和Repair方法具体实现模块
  • Destroy和Repair方法管理模块

下面我们将对其进行一一讲解,不知道大家小板凳准备好了没有。

02 Destroy和Repair方法具体实现

关于Destroy和Repair方法,由三个类组成,分别是AOperator、ADestroyOperator、ARepairOperator。它们之间的继承派生关系如下:

下面对其一一讲解。

2.1 AOperator

这是一个基础父类,它抽象了Destroy和Repair方法共有的一些方法和特征(成绩、权重、名称等等),然后Destroy和Repair方法再各自继承于它,实现自己的功能模块。

成员变量已经注释清楚了,关于protected的一个成员noise噪声模式会在后面讲到。其他的方法也很简单就不做多解释了。

  1class AOperator
  2{
  3private:
  4    //! Total number of calls during the process.
  5    size_t totalNumberOfCalls;
  6
  7    //! Number of calls since the last evaluation.
  8    size_t nbCallsSinceLastEval;
  9
 10    //! score of the operator.
 11    double score;
 12
 13    //! weight of the operator.
 14    double weight;
 15
 16    //! designation of the operator.
 17    std::string operatorName;
 18
 19protected:
 20    //! Indicate if the operator is used in noise mode or not.
 21    bool noise;
 22public:
 23
 24    //! Constructor.
 25    AOperator(std::string name){
 26        operatorName = name;
 27        init();
 28    }
 29
 30    //! Destructor.
 31    virtual ~AOperator(){};
 32
 33    //! Initialize the values of the numbers of calls.
 34    void init()
 35    {
 36        totalNumberOfCalls = 0;
 37        nbCallsSinceLastEval = 0;
 38        score = 0;
 39        weight = 1;
 40    }
 41
 42    //! reset the number of calls since last eval.
 43    void resetNumberOfCalls()
 44    {
 45        nbCallsSinceLastEval = 0;
 46    }
 47
 48    //! Simple getter.
 49    //! \return the total number of calls to the operator since
 50    //! the beginning of the optimization process.
 51    size_t getTotalNumberOfCalls(){return totalNumberOfCalls;};
 52
 53    //! Simple getter.
 54    //! \return the number of calls to this operator since the last
 55    //! evaluation.
 56    size_t getNumberOfCallsSinceLastEvaluation(){return nbCallsSinceLastEval;};
 57
 58    void increaseNumberOfCalls()
 59    {
 60        totalNumberOfCalls++;
 61        nbCallsSinceLastEval++;
 62    }
 63
 64    //! Simple getter.
 65    double getScore() const
 66    {
 67        return score;
 68    }
 69
 70    //! Simple getter.
 71    double getWeight() const
 72    {
 73        return weight;
 74    }
 75
 76    //! resetter.
 77    void resetScore()
 78    {
 79        this->score = 0;
 80    }
 81
 82    //! Simple setter.
 83    void setScore(double s)
 84    {
 85        this->score = s;
 86    }
 87
 88    //! Simple setter.
 89    void setWeight(double weight)
 90    {
 91        this->weight = weight;
 92    }
 93
 94    //! Simple getter.
 95    std::string getName(){return operatorName;};
 96
 97    //! Set noise to true.
 98    void setNoise(){noise=true;};
 99
100    //! Set noise to false.
101    void unsetNoise(){noise=false;};
102
103};

2.2 ADestroyOperator

该类主要是继承于上面的AOperator类,然后再此基础上加上Destroy操作的具体实现。它是一个抽象类,需要在后续的应用中重写Destroy操作的方法。

 1class ADestroyOperator : public AOperator {
 2protected:
 3    //! The minimum destroy size used.
 4    size_t minimunDestroy;
 5    //! The maximum destroy size used.
 6    size_t maximumDestroy;
 7
 8public:
 9    //! Constructor.
10    //! \param mini the minimum destroy size.
11    //! \param maxi the maximum destroy size.
12    //! \param s the name of the destroy operator.
13    ADestroyOperator(size_t mini, size_t maxi, std::string s) : AOperator(s)
14    {
15        minimunDestroy = mini;
16        maximumDestroy = maxi;
17    }
18
19    //! Destructor.
20    virtual ~ADestroyOperator(){};
21
22    //! This function is the one called to destroy a solution.
23    //! \param sol the solution to be destroyed.
24    virtual void destroySolution(ISolution& sol)=0;
25};

2.3 ARepairOperator

同理,也是由AOperator类派生出来并加上Repair自己的实现方法的类。也是一个抽象类,需要在后续的使用中重写Repair方法。

 1class ARepairOperator : public AOperator {
 2
 3public:
 4    ARepairOperator(std::string s) : AOperator(s)
 5    {
 6    }
 7
 8    virtual ~ARepairOperator(){};
 9
10    virtual void repairSolution(ISolution& sol)=0;
11};

03 Destroy和Repair方法管理

对Destroy和Repair方法进行管理的由两个类来实现:AOperatorManager、OperatorManager

其中AOperatorManager是抽象类,只提供接口,OperatorManager继承于AOperatorManager。并对其接口进行实现。然后,接着看代码。

3.1 AOperatorManager

该类抽象了OperatorManager的一些特征,只提供接口因此成员函数都是纯虚函数。相关方法的说明已经注释在代码里面了。

关于保护成员:stats用于保存算法迭代过程中的一些状态量,这个类后续也会讲解的。

 1class AOperatorManager
 2{
 3public:
 4    //! This method selects a destroy operator.
 5    //! \return a destroy operator.
 6    virtual ADestroyOperator& selectDestroyOperator()=0;
 7
 8    //! This method selects a repair operator.
 9    //! \return a repair operator.
10    virtual ARepairOperator& selectRepairOperator()=0;
11
12    virtual void recomputeWeights()=0;
13
14    //! Update the scores of the operators.
15    virtual void updateScores(ADestroyOperator& des, ARepairOperator& rep, ALNS_Iteration_Status& status)=0;
16
17    //! Indicate that the optimization process starts.
18    virtual void startSignal()=0;
19
20    //! Destroy the operators registered to this operator manager.
21    virtual void end()=0;
22
23    //! Simple setter.
24    void setStatistics(Statistics* statistics){stats = statistics;};
25protected:
26    //! A pointer to the instance of the statistics module.
27    Statistics* stats;
28};

3.2 OperatorManager

该类在AOperatorManager基础上也添加了一些自己额外的成员变量和函数方法。具体还是看代码理解吧,挺简单的,没有需要多解释的,我在这多少无益。

 1class OperatorManager: public AOperatorManager {
 2private:
 3    //! The set of repair operators.
 4    std::vector<AOperator*> repairOperators;
 5
 6    //! The set of destroy operators.
 7    std::vector<AOperator*> destroyOperators;
 8
 9    //! The sum of the weights of the repair operators.
10    double sumWeightsRepair;
11
12    //! The sum of the weights of the destroy operators.
13    double sumWeightsDestroy;
14
15    //! The paramaters to be used by the ALNS.
16    ALNS_Parameters* parameters;
17
18    //! Indicate whether or not the next operators to be return
19    //! should be noised or not.
20    bool noise;
21
22    //! A counter that indicates the number of times repair operators with noise have been successfull
23    double performanceRepairOperatorsWithNoise;
24    //! A counter that indicates the number of times repair operators without noise have been successfull
25    double performanceRepairOperatorsWithoutNoise;
26
27
28    //! Use a roulette wheel to select an operator in a vector of operators.
29    //! \return the selected operator.
30    AOperator& selectOperator(std::vector<AOperator*>& vecOp, double sumW);
31
32    //! Recompute the weight of an operator.
33    void recomputeWeight(AOperator& op, double& sumW);
34public:
35    //! Constructor
36    //! \param param the parameters to be used.
37    OperatorManager(ALNS_Parameters& param);
38
39    //! Destructor.
40    virtual ~OperatorManager();
41
42    //! This function recompute the weights of every operator managed by this
43    //! manager.
44    void recomputeWeights();
45
46    //! This method selects a destroy operator.
47    //! \return a destroy operator.
48    ADestroyOperator& selectDestroyOperator();
49
50    //! This method selects a repair operator.
51    //! \return a repair operator.
52    ARepairOperator& selectRepairOperator();
53
54    //! This method adds a repair operator to the list
55    //! of repair operator managed by this manager.
56    //! \param repairOperator the repair operator to be added.
57    void addRepairOperator(ARepairOperator& repairOperator);
58
59    //! This method adds a destroy operator to the list
60    //! of destroy operator managed by this manager.
61    //! \param destroyOperator the destroy operator to be added.
62    void addDestroyOperator(ADestroyOperator& destroyOperator);
63
64    //! This method run some sanity checks to ensure that it is possible
65    //! to "safely" use this manager within the ALNS.
66    void sanityChecks();
67
68    //! Update the scores of the operators.
69    virtual void updateScores(ADestroyOperator& des, ARepairOperator& rep, ALNS_Iteration_Status& status);
70
71    //! Indicate that the optimization process starts.
72    virtual void startSignal();
73
74    //! Destroy all the operators registered to this operator.
75    void end();
76};

上面是该类的.h文件,关于其中某些函数方法的实现,小编下面挑一些来重点给大家讲讲,那些以小编的脑瓜子都能理解的代码就省略了,大家应该都能懂……

04 OperatorManager具体实现

又到了一大波代码时间,来吧来吧,小板凳准备好,要开始啦~

4.1 OperatorManager::recomputeWeight()

重新计算单个操作的权重。其有两个参数AOperator& op, double& sumW。

其中 op是要重新计算权重的repair或者destroy方法,sumW是其对应集合的权重总和。这里只讲一个新权重的计算方式就行:

其中:

Rho为设定的[0, 1]之间的参数。

PrevWeight表示旧的权重。

nbCalls表示在上一次自上一次更新完权重到现在该方法被调用的次数。

timeSegmentsIt表示迭代多少次需要重新计算一下权重的迭代次数。

currentScore表示旧的成绩。

理解了这些就很easy了。

 1void OperatorManager::recomputeWeight(AOperator& op, double& sumW)
 2{
 3    double prevWeight = op.getWeight();
 4    sumW -= prevWeight;
 5    double currentScore = op.getScore();
 6    size_t nbCalls = op.getNumberOfCallsSinceLastEvaluation();
 7    double newWeight = (1-parameters->getRho())*prevWeight + parameters->getRho()*(static_cast<double>(nbCalls)/static_cast<double>(parameters->getTimeSegmentsIt()))*currentScore;
 8    // We ensure that the weight is within the bounds.
 9    if(newWeight > parameters->getMaximumWeight())
10    {
11        newWeight = parameters->getMaximumWeight();
12    }
13    if(newWeight < parameters->getMinimumWeight())
14    {
15        newWeight = parameters->getMinimumWeight();
16    }
17
18    sumW += newWeight;
19    op.setWeight(newWeight);
20    op.resetScore();
21    op.resetNumberOfCalls();
22}

值得注意的是还有一个OperatorManager::recomputeWeights()成员函数是用于重新计算repair或者destroy方法集合的。

它的实现主要也还是调用OperatorManager::recomputeWeight(AOperator& op, double& sumW)方法来实现的。

4.2 OperatorManager::selectOperator(...)

相信了解过遗传算法轮盘赌实现过程的小伙伴对这里都不会陌生,当然,并不是说权重大的方法一定会被选中,只是被选中的可能性会大而已。

具体过程是先生成一个在0到sumWeight之间的中间权重randomWeightPos ,然后从第一个方法开始用变量cumulSum进行权重累加,直到cumulSum>=randomWeightPos 为止,那么停止累加时最后这个方法就是幸运儿了。

 1AOperator& OperatorManager::selectOperator(std::vector<AOperator*>& vecOp, double sumW)
 2{
 3    double randomVal = static_cast<double>(rand())/static_cast<double>(RAND_MAX);
 4    double randomWeightPos = randomVal*sumW;
 5    double cumulSum = 0;
 6    for(size_t i = 0; i < vecOp.size(); i++)
 7    {
 8        cumulSum += vecOp[i]->getWeight();
 9        if(cumulSum >= randomWeightPos)
10        {
11            if(noise)
12            {
13                vecOp[i]->setNoise();
14            }
15            else
16            {
17                vecOp[i]->unsetNoise();
18            }
19            vecOp[i]->increaseNumberOfCalls();
20            return *(vecOp[i]);
21        }
22    }
23    assert(false);
24    return *(vecOp.back());
25}

4.3 OperatorManager::updateScores(...)

该成员函数用来更新各个Destroy和Repair方法的成绩。参数是Destroy和Repair方法的集合,以及ALNS迭代过程中的各种状态信息

便于说明下面用rScore和dScore分别代表Repair和Destroy方法的成绩。具体实现如下:

1) 如果找到新的最优解,rScore+=Sigma1,dScore+=Sigma1。其中Sigma1是设定参数。

2) 如果当前解得到改进,rScore+=Sigma2,dScore+=Sigma2其中Sigma2是设定参数。

3) 如果当前解没有得到改进 and 当前解是之前没有出现过的 and 当前解被接受作为新的解了,rScore+=Sigma3,dScore+=Sigma3。其中Sigma3是设定参数。

 1void OperatorManager::updateScores(ADestroyOperator& des, ARepairOperator& rep, ALNS_Iteration_Status& status)
 2{
 3    if(status.getNewBestSolution() == ALNS_Iteration_Status::TRUE)
 4    {
 5        rep.setScore(rep.getScore()+parameters->getSigma1());
 6        des.setScore(des.getScore()+parameters->getSigma1());
 7        performanceRepairOperatorsWithNoise += 1;
 8        performanceRepairOperatorsWithoutNoise += 1;
 9    }
10
11    if(status.getImproveCurrentSolution() == ALNS_Iteration_Status::TRUE)
12    {
13        rep.setScore(rep.getScore()+parameters->getSigma2());
14        des.setScore(des.getScore()+parameters->getSigma2());
15        performanceRepairOperatorsWithNoise += 1;
16        performanceRepairOperatorsWithoutNoise += 1;
17    }
18
19    if(status.getImproveCurrentSolution() == ALNS_Iteration_Status::FALSE
20            && status.getAcceptedAsCurrentSolution() == ALNS_Iteration_Status::TRUE
21            && status.getAlreadyKnownSolution() == ALNS_Iteration_Status::FALSE)
22    {
23        rep.setScore(rep.getScore()+parameters->getSigma3());
24        des.setScore(des.getScore()+parameters->getSigma3());
25        performanceRepairOperatorsWithNoise += 1;
26        performanceRepairOperatorsWithoutNoise += 1;
27    }
28    /* OLD VERSION */
29    /*
30    if(parameters->getNoise())
31    {
32        double randNoise = static_cast<double>(rand())/RAND_MAX;
33        noise = (randNoise<parameters->getProbabilityOfNoise());
34    }
35    */
36    /* NEW VERSION */
37
38    if(parameters->getNoise())
39    {
40        double performanceRepairOperatorsGlobal = 0;
41        performanceRepairOperatorsGlobal += performanceRepairOperatorsWithNoise;
42        performanceRepairOperatorsGlobal += performanceRepairOperatorsWithoutNoise;
43
44        double randomVal = static_cast<double>(rand())/RAND_MAX;
45        double randomWeightPos = randomVal*performanceRepairOperatorsGlobal;
46        noise = (randomWeightPos < performanceRepairOperatorsGlobal);
47    }
48}

至此,差不过难点都讲完了,不知道你萌都understand了吗?

05 小结

好了,以上就是今天的代码内容,别急着走开哦,后面还有好多好多的内容没讲呢。

这是个天坑,希望大家和小编一起努力,共同填完它。哈哈,谢谢各位的至此。对了,你萌不会不点赞就走了吧?

END

【如对代码有疑问,可联系小编,可以提供有偿辅导服务】

【有偿辅导纯属个人行为,与团队无关】

精彩文章推荐

干货 | 变邻域搜索算法(VNS)求解TSP(附C++详细代码及注释)

干货 | 变邻域搜索算法(Variable Neighborhood Search,VNS)超详细一看就懂

干货 | 遗传算法(Genetic Algorithm) Java 详细代码及注释

干货 | 遗传算法(Genetic Algorithm) (附代码及注释)

干货|迭代局部搜索算法(Iterated local search)探幽(附C++代码及注释)

干货 | 用模拟退火(SA, Simulated Annealing)算法解决旅行商问题

---The End---

文案 && 编辑:邓发珩

审稿:庄浩城

指导老师: 秦时明岳(华中科技大学管理学院)

如对文中内容有疑问,欢迎交流。PS:部分资料来自网络。

如有需求,可以联系:

秦虎老师(professor.qin@qq.com)

邓发珩 (华中科技大学管理学院本科二年级:2638512393@qq.com、个人公众号:程序猿声)

庄浩城(华中科技大学管理学院本科三年级:1604088003@qq.com)

本文参与 腾讯云自媒体分享计划,分享自微信公众号。
原始发表:2019-02-22,如有侵权请联系 cloudcommunity@tencent.com 删除

本文分享自 数据魔术师 微信公众号,前往查看

如有侵权,请联系 cloudcommunity@tencent.com 删除。

本文参与 腾讯云自媒体分享计划  ,欢迎热爱写作的你一起参与!

评论
登录后参与评论
0 条评论
热度
最新
推荐阅读
领券
问题归档专栏文章快讯文章归档关键词归档开发者手册归档开发者手册 Section 归档