Can Automated Impact Analysis Technique Help Predicting Decaying Modules?
0 likes10,785 views
This study examines whether impact analysis (IA) can help predict decaying modules in software projects. The researchers used IR-based IA to connect bug reports to source code modules. They found that using IA improved prediction accuracy over a baseline approach using only code quality metrics. Further improvements could be made by increasing IA's recall through adding true positive connections and prioritizing recall over precision. The goal is to help developers proactively refactor decaying modules before they become severely degraded.
![CanAutomatedImpactAnalysis
TechniqueHelpPredicting
DecayingModules?
Tokyo Institute of
Technology
Natthawute
Sae-Lim
Shinpei
Hayashi
Motoshi
Saeki
[ Smelling things by Tjarko Busink: https://www.flickr.com/photos/sjekkiebunzing/13009723845/ ]](https://image.slidesharecdn.com/canautomatedimpactanalysistechniquehelppredictingdecayingmodules-200323094011/85/Can-Automated-Impact-Analysis-Technique-Help-Predicting-Decaying-Modules-1-320.jpg)
![[ Smelling things by Tjarko Busink: https://www.flickr.com/photos/sjekkiebunzing/13009723845/ ]
Bad Smell](https://image.slidesharecdn.com/canautomatedimpactanalysistechniquehelppredictingdecayingmodules-200323094011/85/Can-Automated-Impact-Analysis-Technique-Help-Predicting-Decaying-Modules-2-320.jpg)
![5
approx. 19% of modules
were decaying modules
in each release
’’
Sae-Lim et al.,“Toward Proactive Refactoring:An Exploratory Study on Decaying Modules”, Proc. IWoR 2019.
[ 3/4" 285HV Grid 1749 by Peter Kemmer: https://www.flickr.com/photos/pkmousie/2225340795/ ]
decaying modules are
more likely to get decayed
in later releases
’’](https://image.slidesharecdn.com/canautomatedimpactanalysistechniquehelppredictingdecayingmodules-200323094011/85/Can-Automated-Impact-Analysis-Technique-Help-Predicting-Decaying-Modules-5-320.jpg)
![Research Questions
RQ1:
Can IR-based impact analysis
improve prediction performance?
7[ protractors by Dean Hochman: https://www.flickr.com/photos/deanhochman/14481958642/ ]
RQ2:
How can we further improve
prediction performance?](https://image.slidesharecdn.com/canautomatedimpactanalysistechniquehelppredictingdecayingmodules-200323094011/85/Can-Automated-Impact-Analysis-Technique-Help-Predicting-Decaying-Modules-7-320.jpg)
![CanAutomated
ImpactAnalysis
TechniqueHelp
Predicting
DecayingModules?
[ Smelling things by Tjarko Busink: https://www.flickr.com/photos/sjekkiebunzing/13009723845/ ]
Can IR-Based IA improve
prediction performance?
11
RQ
1
If IA is perfect, accuracy becomes very high
Future direction: Improving IA
Non
Smell
Decaying Modules:
Modules going to be smelly
3
Smell
Module
Decaying
Index
0
1
Smell
Severity
Decaying modules ⇒ whose smell degree
has increased from the previous release
Time
How can we further
improve the performance?
16
RQ
2
HiveDerby
Adding TPs (recall!) should be done first
AccumuloAmbari
Precision
Recall
IA accuracy↑ leads to Performance↑](https://image.slidesharecdn.com/canautomatedimpactanalysistechniquehelppredictingdecayingmodules-200323094011/85/Can-Automated-Impact-Analysis-Technique-Help-Predicting-Decaying-Modules-16-320.jpg)
Can Automated Impact Analysis Technique Help Predicting Decaying Modules?
- 1. CanAutomatedImpactAnalysis TechniqueHelpPredicting DecayingModules? Tokyo Institute of Technology Natthawute Sae-Lim Shinpei Hayashi Motoshi Saeki [ Smelling things by Tjarko Busink: https://www.flickr.com/photos/sjekkiebunzing/13009723845/ ]
- 2. [ Smelling things by Tjarko Busink: https://www.flickr.com/photos/sjekkiebunzing/13009723845/ ] Bad Smell
- 3. Code Smell 3 An indicator of a design flaw/problem in the source code One of the factors that cause technical debt Increases code component’s fault-proneness Detector Code smells Refactoring Class A { __________ __________ __________ __________ } Source code How to prevent?
- 4. Non Smell Decaying Modules: Modules going to be smelly 4 Smell Module Decaying Index 0 1 Smell Severity Decaying modules ⇒ whose smell degree has increased from the previous release Time
- 5. 5 approx. 19% of modules were decaying modules in each release ’’ Sae-Lim et al.,“Toward Proactive Refactoring:An Exploratory Study on Decaying Modules”, Proc. IWoR 2019. [ 3/4" 285HV Grid 1749 by Peter Kemmer: https://www.flickr.com/photos/pkmousie/2225340795/ ] decaying modules are more likely to get decayed in later releases ’’
- 6. Decaying Module Prediction l Existing approach: code quality metric-based l Use of task context – Using IR-based impact analysis (IA) to connect change descriptions to modules may contribute to improving the prediction performance 6 Source code Source code Bug 123 When click… Changes Bug 123 When click… Bug 123 When click… Ver 1.0 Ver 2.0 Implementation
- 7. Research Questions RQ1: Can IR-based impact analysis improve prediction performance? 7[ protractors by Dean Hochman: https://www.flickr.com/photos/deanhochman/14481958642/ ] RQ2: How can we further improve prediction performance?
- 8. Baseline 8 All data Training data Static Analyzer Detector Model Construction 1._______ 2._______ 3._______ 4._______ Current RecentHistorical Code quality Decaying modules 1… 2… 3… Predicted results Test data Static Analyzer Trained model Vn Vn+1 1._______ 2._______ 3._______ 4._______ Current RecentHistorical Code quality Pantiuchina, et al.:“Towards just-in-time refactoring recommenders”, Proc. ICPC 2018.
- 9. Baseline + Impact Analysis 10 Bug 123 When click… Bug 123 When click… Bug 123 When click… Change descriptions Impact Analysis Main() xxx;; Source code Estimated module list Bug 123 When click… Bug 123 When click… Bug 123 When click… Change descriptions Impact Analysis Main() xxx;; Source code Estimated module list Model Construction 1._______ 2._______ 3._______ 4._______ Current RecentHistorical Code quality Decaying modules 1… 2… 3… Predicted results Trained model 1._______ 2._______ 3._______ 4._______ Current RecentHistorical Code quality
- 10. Dataset Target smell: God Class 11 Project Period (years) # Releases Accumulo 6 30 Ambari 6 33 Hive 13 28 Derby 7 33
- 11. Can IR-based IA improve prediction performance? 12 RQ 1
- 12. Can IR-based IA improve prediction performance? 13 RQ 1 If IA is perfect, accuracy becomes very high. Future direction: Improving IA
- 13. Delete false positives (FPs) Insert true positives (TPs) How can we further improve the performance? 14 RQ 2 Change description Bug 123 When click… 1Oracle: 2 3 4 5 IA 1 2 1 2 1 2 3 Result: Simulating IA results of different accuracy:
- 14. How can we further improve the performance? 15 RQ 2 FP Dec. Ratio TP Inc. Ratio 0 1 1 0 Accumulo
- 15. How can we further improve the performance? 16 RQ 2 HiveDerby Adding TPs (recall!) should be done first AccumuloAmbari Precision Recall IA accuracy↑ leads to Performance↑
- 16. CanAutomated ImpactAnalysis TechniqueHelp Predicting DecayingModules? [ Smelling things by Tjarko Busink: https://www.flickr.com/photos/sjekkiebunzing/13009723845/ ] Can IR-Based IA improve prediction performance? 11 RQ 1 If IA is perfect, accuracy becomes very high Future direction: Improving IA Non Smell Decaying Modules: Modules going to be smelly 3 Smell Module Decaying Index 0 1 Smell Severity Decaying modules ⇒ whose smell degree has increased from the previous release Time How can we further improve the performance? 16 RQ 2 HiveDerby Adding TPs (recall!) should be done first AccumuloAmbari Precision Recall IA accuracy↑ leads to Performance↑