運用模型組合方法於開放源碼軟體規模變動之分析

Using Model Combinatorial Methods to Analyze Program Size Variation in Open-Source Software

黃詩閔(Shih-Min Huang)

指導教授 ： 黃慶育

國立清華大學/電機資訊學院/資訊系統與應用研究所/碩士(2013年)

摘要

在軟體開發的過程中，軟體大小的測量是一個非常重要且會影響專案是否會成功的關鍵因素之一。一般來說，軟體大小的衡量可分為，軟體行碼數，功能點的多寡，以及軟體複雜度。在實務上，因為軟體行碼數的易於計算和廣泛使用於各種軟體測量領域，大部分的軟體開發工程師在專案完成時還是以軟體行碼數當作測量軟體大小的尺度。在過去有一些研究發現指出，Lognormal distribution 和 Double-pareto distribution 模型均可以用來描述專案裡軟體大小分佈情形。然而本文切換另外一種角度探討軟體行碼數的變動率的分佈情形，我們觀察到軟體規模的變動率和經濟學的利率變動情形有相似的地方，所以我們使用衡量利率變動情形的Laplace distribution，Normal distribution 和 Asymmetric laplace distribution來衡量軟體大小的變動率。
　　目前許多的研究都分別指出沒有最好的模型適用在所有的情況，我們利用一種基於貝氏定理權重決定法(BIWDA)為修改的模型組合方法: Modified BIWDA(MBIWDA)來塑造軟體大小變動率分佈。研究資料來源於Apache，Ubuntu和Samba Server來確認利用MBIWDA的組合模型的軟體大小變動率分佈之預測能力。與其他相關模型組合方法比較後，本研究結果指出利用MBIWDA的組合模型能夠在大部份情況下提供不錯的預測能力。並且，我們也利用分量迴歸分析來探討影響軟體大小的變動因素，研究發現在軟體早期階段軟體大小的變動會被嚴重等級較高的錯誤所影響。這些發現對於分析開放源碼的軟體大小變動率都有很大的啟發。

關鍵字

軟體大小估計 ； 程式碼長度 ； 組合模型 ； 分量迴歸

並列摘要

As one of the most important internal attributes of software systems, the estimation of software size is crucial to project success. Typically, software size can be described by the length, functionality, or complexity of the file, but in practice many people still use lines of code (LOC) as a measure of software size since LOC is widely used and can be easily measured upon project completion.
In this paper, we used a linear combination model with modified Bayesian inference weight decision approach (MBIWDA) to analyze the size distribution and software size-change rate of Open-Source Software (OSS). Furthermore, we investigated the factors that influenced the software size-change rate using the quantile regression (QR) model. Experiments were conducted using real data of several OSS projects, and evaluation results showed that the linear combination model with the MBIWDA had a outstanding capability of fitting the distribution of the software size-change rate. Finally, the analysis of QR demonstrated that faults with higher severity had an impact on LOC changes in the early stage. These findings offer an alternative view and reveal different issues of software sizing.

並列關鍵字

Size estimation ； Code length ； Combination Model ； Quantile Regression

參考文獻

[1]. R. Sharma and R. S. Chhillar, “Novel Approach to Software Metrics,” International Journal of Soft Computing and Engineering, Vol 2, Issue. 3 Jul 2012

[2]. I. Herraiz, D. M. German and A. E. Hassan, “On the Distribution of Source Code File Sizes,” Proceedings of the 6th International Conference on Software and Data Technologies (ICSOFT 2011) pp.18-21, Seville, Spain, July 2011.

[3]. G. Concas, M. Marchesi, S. Pinna, and N. Serra, “Power-laws in a Large Oriented Software System,” IEEE Trans. on Software Engineering, Vol. 33, Issue. 10, pp. 687-708, Oct 2007.

[5]. K. Samuel, T. J. Kozubowski, and K. Podgorski, The Laplace Distribution and Generalizations: A Revisit with Applications to Communications, Economics, Engineering, and Finance, No. 183. Springer, 2001.

[6]. S. Gennady, and M. S. Taqqu, “Stable Non-Gaussian Random Processes,” Econometric Theory 13 pp. 133-142, 1997.