面向对象软件度量因子的幂律分布研究
|
摘要
结合幂律分布的研究方法对面向对象软件度量的结果进行研究,探讨不同结构层次软件度量中出现的长尾分布现象。基于4个大型Java开源项目,分别对类代码长度、类属性个数和类方法数这3个影响软件质量的主要度量因子进行度量。度量结果拟合幂律分布,采用极大似然估计法对拟合的幂律分布形状参数进行估计,并使用基于拔靴法的假设检验方法评估拟合效果。通过对3个度量因子统计结果研究发现,幂律分布存在于大部分统计结果的尾部。利用度量因子拟合的幂律分布函数,在互补累积分布图中找到偏离拟合幂律分布函数较远的类,可以在这些类中发现代码质量问题。
In this paper,the method of power-law distribution was used to study the results of object-oriented software metrics,and the long tailed distribution in different levels of software metrics was discussed. Based on four large Java open source projects,three major metrics affecting software quality were measured,such as class code length,class attribute number,and class method number. First,we utilized the measurements to fit the power-law distribution,and the maximum likelihood estimation method was used to estimate the shape parameter of the fitted power-law distribution.Then the hypothesis test method based on the bootstrap method was used to evaluate the fitting effect. Moreover,it was found that the power-law distribution existed at the tail of most statistical results by studying the statistical results of three metric factors. After fitting the power-law distribution function to the statistical result,it will lead to code quality issues in CCDF plots if we find that the measurement result of deviation from the fitting function is large.
In this paper,the method of power-law distribution was used to study the results of object-oriented software metrics,and the long tailed distribution in different levels of software metrics was discussed. Based on four large Java open source projects,three major metrics affecting software quality were measured,such as class code length,class attribute number,and class method number. First,we utilized the measurements to fit the power-law distribution,and the maximum likelihood estimation method was used to estimate the shape parameter of the fitted power-law distribution.Then the hypothesis test method based on the bootstrap method was used to evaluate the fitting effect. Moreover,it was found that the power-law distribution existed at the tail of most statistical results by studying the statistical results of three metric factors. After fitting the power-law distribution function to the statistical result,it will lead to code quality issues in CCDF plots if we find that the measurement result of deviation from the fitting function is large.
引文
[1]Boehm B.Industrial Software Metrics Top 10 List[J].IEEESoftware,1987,4:84-85.
[2]Clauset A,Shalizi C R,Newman M E J.Power-Law Distributions in Empirical Data[J].Siam Review,2014,51(4):661-703.
[3]马皖王莹,陈林,陈芝菲,等.基于复杂网络分析的软件高危缺陷评估方法[J].计算机科学与探索,2014,8(8):956-965.
[4]韩明畅,李德毅,刘常昱,等.软件中的网络化特征及其对软件质量的贡献[J].计算机工程与应用,2006,42(20):29-31.
[5]Valverde S,Sole R V.Hierarchical Small Worlds in Software Architecture[J].Dynamics of Continuous Discrete&Impulsive Systems,2007,14.
[6]Valverde S,Cancho R F I,Sole R V.Scale-free Networks from Optimal Design[J].Epl,2002,60(4):512-517.
[7]Louridas P,Spinellis D,Vlachos V.Power Laws in Software[J].ACM Transactions on Software Engineering&Methodology,2008,18(1):617-632.
[8]Tempero E,Anslow C,Dietrich J,et al.The Qualitas Corpus:A Curated Collection of Java Code for Empirical Studies[C]//Asia Pacific Software Engineering Conference.IEEEComputer Society,2010:336-345.
[9]Liu Hongtao.Perfume:A code quality assessment platform based on machine learning[EB/OL].https://github.com/Liu Hongtao/Perfume.
[10]Gillespie C S.Fitting Heavy Tailed Distributions:The poweR-law Package[J].Journal of Statistical Software,2015,64(2).
[11]Concas G,Marchesi M,Pinna S,et al.Power-Laws in a Large Object-Oriented Software System[J].IEEE Transactions on Software Engineering,2007,33(10):687-708.
[2]Clauset A,Shalizi C R,Newman M E J.Power-Law Distributions in Empirical Data[J].Siam Review,2014,51(4):661-703.
[3]马皖王莹,陈林,陈芝菲,等.基于复杂网络分析的软件高危缺陷评估方法[J].计算机科学与探索,2014,8(8):956-965.
[4]韩明畅,李德毅,刘常昱,等.软件中的网络化特征及其对软件质量的贡献[J].计算机工程与应用,2006,42(20):29-31.
[5]Valverde S,Sole R V.Hierarchical Small Worlds in Software Architecture[J].Dynamics of Continuous Discrete&Impulsive Systems,2007,14.
[6]Valverde S,Cancho R F I,Sole R V.Scale-free Networks from Optimal Design[J].Epl,2002,60(4):512-517.
[7]Louridas P,Spinellis D,Vlachos V.Power Laws in Software[J].ACM Transactions on Software Engineering&Methodology,2008,18(1):617-632.
[8]Tempero E,Anslow C,Dietrich J,et al.The Qualitas Corpus:A Curated Collection of Java Code for Empirical Studies[C]//Asia Pacific Software Engineering Conference.IEEEComputer Society,2010:336-345.
[9]Liu Hongtao.Perfume:A code quality assessment platform based on machine learning[EB/OL].https://github.com/Liu Hongtao/Perfume.
[10]Gillespie C S.Fitting Heavy Tailed Distributions:The poweR-law Package[J].Journal of Statistical Software,2015,64(2).
[11]Concas G,Marchesi M,Pinna S,et al.Power-Laws in a Large Object-Oriented Software System[J].IEEE Transactions on Software Engineering,2007,33(10):687-708.