【摘要】：定位軟件中存在的故障或者說找到軟件中錯誤語句所在的行數(shù)是軟件開發(fā)周期中最耗時的部分,因此研究自動化故障定位技術(shù)或者輔助故障定位技術(shù)都是非常有必要的。現(xiàn)有的故障定位技術(shù)中以基于程序譜的技術(shù)在實際作用中效率最高。然而該技術(shù)存在著比較明顯的缺陷:(1)對于所有程序,采用固定的懷疑度公式計算程序?qū)嶓w的懷疑度,然而對于特定的程序,受到測試用例以及開發(fā)語言等各種客觀因素影響,可能出錯方式與其他程序是不同的。(2)出現(xiàn)故障的語句往往會在程序運行的過程中影響后面的語句,但該技術(shù)沒有考慮程序?qū)嶓w之間的這種關(guān)聯(lián),因此無法正確的評估某些語句的懷疑度值,進而導(dǎo)致定位效果不佳。針對以上問題本文提出采用核嶺回歸對程序?qū)嶓w與測試用例運行結(jié)果之間的關(guān)系構(gòu)建學(xué)習(xí)模型,嶺回歸方法是一種能夠解決特征之間存在復(fù)共線性關(guān)系的機器學(xué)習(xí)方法,因此可以對每一個程序構(gòu)建自己獨特的程序?qū)嶓w懷疑度計算過程,不失一般性本文將語句作為程序?qū)嶓w,將語句在執(zhí)行時的覆蓋次數(shù)作為特征,將每一個測試用例的運行結(jié)果作為目標值,使用每一次測試用例運行時程序的語句覆蓋情況以及目標結(jié)果構(gòu)建訓(xùn)練矩陣,最終通過使用核嶺回歸進行樣本訓(xùn)練得到的擬合曲線以及虛擬向量來評估每一項程序?qū)嶓w包含故障的可能性。由于嶺回歸方法中目標結(jié)果為二值的情況并不能完全適應(yīng)于軟件故障定位的要求,因此本文還提出了對測試用例運行結(jié)果的平滑過程,以優(yōu)化擬合曲線的構(gòu)建。最后,本文設(shè)計了詳細的實驗框架,建立了自動化的故障定位流程,按照現(xiàn)有故障定位方法的評估標準使用通用測試集驗證了基于嶺回歸構(gòu)建實體懷疑度來評估程序?qū)嶓w包含故障可能性大小方法的有效性。
[Abstract]:It is the most time-consuming part of the software development cycle to locate the faults in the software or to find the number of lines where the wrong statements are located, so it is necessary to study the automatic fault location technology or the auxiliary fault location technology. Among the existing fault location techniques, the one based on program spectrum is the most efficient in practice. However, the technique has some obvious defects: (1) for all programs, a fixed degree of doubt formula is used to calculate the degree of doubt of a program entity. However, for a particular program, it is affected by various objective factors, such as test cases and development languages, etc. Possible errors are different from other programs. (2) statements that fail tend to affect subsequent statements while the program is running, but the technique does not take into account this association between program entities, Therefore, the doubt value of some statements can not be correctly evaluated, which leads to poor positioning effect. In order to solve the above problems, this paper proposes a learning model based on kernel ridge regression, which can solve the complex collinear relationship between features and the relationship between program entities and test case results. Therefore, each program can be constructed its own unique program entity doubt calculation process, without losing the general statement as a program entity, the statement in the execution of the coverage of the number of times as a feature, The result of each test case is taken as the target value, and the training matrix is constructed using the statement coverage of the program and the target result of each test case run. Finally, the probability of each program entity containing the fault is evaluated by using the fitting curve and the virtual vector obtained by using the kernel ridge regression to train the samples. Because the target result is binary in the ridge regression method, it can not meet the requirement of software fault location. Therefore, the smooth process of test case running results is proposed in this paper to optimize the construction of fitting curve. Finally, the detailed experimental framework is designed, and the automatic fault location flow is established. According to the evaluation criteria of the existing fault location methods, the validity of the method for evaluating the probability of the program entities containing faults is verified by using the universal test set to construct the degree of entity suspicion based on the ridge regression.
【學(xué)位授予單位】：浙江大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP311.5

【參考文獻】

相關(guān)期刊論文 前1條

1 曹鶴玲;姜淑娟;鞠小林;;軟件錯誤定位研究綜述[J];計算機科學(xué);2014年02期

，

本文編號：2289947