Developer-Related Factors in Change Prediction: An Empirical Assessment
0 likes227 views
The document discusses an empirical assessment of developer-related factors in predicting code changes, emphasizing the importance of understanding which code elements exhibit higher change-proneness. It evaluates the performance of developer-based prediction models and compares them to existing techniques, showcasing their complementarity and potential for improved accuracy. Additionally, the study examines modeling factors such as the entropy of changes and developer behavior to enhance change prediction in software development.
Developer-Related Factors in Change Prediction: An Empirical Assessment
- 1. Developer-Related Factors in Change Prediction:An Empirical Assessment May 22 and 23 Buenos Aires - Argentina Gemma Catolino*, Fabio Palomba°, Andrea De Lucia*, Filomena Ferrucci*, Andy Zaidman° *University of Salerno, °Delft University of Technology
- 3. Knowing in advancethe code elements potentially exhibiting a higher change-proneness is crucialfor developers
- 4. CHANGE PREDICTION Effective technique to identify the classes more prone to be modified in the future
- 7. PREDICTION CHANGE Factors influencing the phenomenon Product and evolution metrics influencing the phenomenon LITERATURE
- 8. PREDICTION CHANGE Factors influencing the phenomenon Devising Prediction Models Product and evolution metrics influencing the phenomenon LITERATURE
- 10. non-focused developers could introduce maintainability pitfalls that lead to increase the change-proneness of the modified classes
- 11. Developer-factor can be a useful source of information to predict classes more likely to be changed in the future
- 12. BASIC CODE CHANGE MODEL Entropy of changes applied by developers in a given time period Developer-related factors
- 13. DEVELOPER CHANGE BASED MODEL Developers structural and semantic scattering Developer-related factors BASIC CODE CHANGE MODEL Entropy of changes applied by developers in a given time period
- 14. DEVELOPER MODEL Number of developers who modified a code component in a give time period Developer-related factors DEVELOPER CHANGE BASED MODEL Developers structural and semantic scattering BASIC CODE CHANGE MODEL Entropy of changes applied by developers in a given time period
- 15. Developer-related factors DEVELOPER MODEL Number of developers who modified a code component in a give time period DEVELOPER CHANGE BASED MODEL Developers structural and semantic scattering BASIC CODE CHANGE MODEL Entropy of changes applied by developers in a given time period
- 16. STUDY EMPIRICAL
- 17. Performanceof developer-based prediction models GOAL
- 18. Performance comparedwith the state of the art and complementarity GOAL Performanceof developer-based prediction models
- 19. GOAL Performance comparedwith the state of the art and complementarity Elish model Zhou model EM CM
- 20. 10 SYSTEMS
- 21. ORACLE
- 22. DISTRIBUTION OF CHANGES OF ALL CLASSES OF A SYSTEM 0 N ORACLE
- 23. DISTRIBUTION OF CHANGES OF ALL CLASSES OF A SYSTEM 0 N C1 ORACLE
- 24. DISTRIBUTION OF CHANGES OF ALL CLASSES OF A SYSTEM 0 N C1 t ORACLE
- 25. DISTRIBUTION OF CHANGES OF ALL CLASSES OF A SYSTEM 0 N NUMBER OF CHANGES MEDIAN >= CHANGE PRONE ORACLE
- 26. DISTRIBUTION OF CHANGES OF ALL CLASSES OF A SYSTEM 0 N NUMBER OF CHANGES MEDIAN < NON-CHANGE PRONE ORACLE
- 30. EVALUATION METRICS OVERLAP Mi Mj % of true positives correctly identified by both models
- 31. EVALUATION METRICS OVERLAP Mi Mj % of true positives correctly identified by both models % change-prone classes correctly classified by Mi only and missed by Mj
- 33. TIME WINDOWS CLASSIFIER VALIDATION TECHNIQUES LOGISTIC REGRESSION
- 35. TIME WINDOWS CLASSIFIER VALIDATION TECHNIQUES TW1 HISTORY OF THE SYSTEM DIVIDED IN TIME PERIOD OF THREE MONTHS TW2 TW3
- 36. TIME WINDOWS CLASSIFIER VALIDATION TECHNIQUES HISTORY OF THE SYSTEM DIVIDED IN TIME PERIOD OF THREE MONTHS WE TRAIN THE MODEL ON TW1 TW2 TW3
- 37. TIME WINDOWS CLASSIFIER VALIDATION TECHNIQUES HISTORY OF THE SYSTEM DIVIDED IN TIME PERIOD OF THREE MONTHS TO TEST THE PERFORMANCE OF TW1 TW2 TW3
- 38. TIME WINDOWS CLASSIFIER VALIDATION TECHNIQUES HISTORY OF THE SYSTEM DIVIDED IN TIME PERIOD OF THREE MONTHS TW1 TW2 TW3
- 39. TIME WINDOWS CLASSIFIER VALIDATION TECHNIQUES HISTORY OF THE SYSTEM DIVIDED IN TIME PERIOD OF THREE MONTHS TW1 TW2 TW3
- 40. RESULTS
- 41. 0 25 50 75 100 BCCM DCBM DM Performanceof developer-based prediction models F-MEASUREOVERALL
- 42. class org.gjt.sp.BufferHistory JEDIT AUGUST OCTOBER 19 CHANGES BY ONE DEVELOPER!!!! 2009
- 43. DEVELOPER MODEL NON-CHANGE PRONE JEDIT AUGUST OCTOBER 19 CHANGES BY ONE DEVELOPER!!!! class org.gjt.sp.BufferHistory
- 44. 36 MODIFICATIONS OVER DIFFERENT PACKAGE!!! High level of both semantic and structural scattering! AUGUST OCTOBER JEDIT class org.gjt.sp.BufferHistory
- 45. 36 MODIFICATIONS OVER DIFFERENT PACKAGE!!! AUGUST OCTOBER JEDIT class org.gjt.sp.BufferHistory DEVELOPER CHANGE BASED MODEL CHANGE PRONE
- 46. 0 25 50 75 100 BCCM DCBM DM CM EM Performanceof developer-based prediction models F-MEASUREOVERALL
- 47. 0 25 50 75 100 BCCM DCBM DM CM EM 0 25 50 75 100 BCCM DCBM DM CM EM Performance comparedwith the state of the art and complementarity F-MEASUREOVERALL
- 49. svg.io.SVGFigureFactory RESPONSIBLE FOR PERFORMING THE MAIN FUNCTIONALITY OF ENTIRE THE PROJECT! JHOTDRAW FREQUENTLY MODIFIED!
- 50. svg.io.SVGFigureFactory RESPONSIBLE FOR PERFORMING THE MAIN FUNCTIONALITY OF ENTIRE THE PROJECT! JHOTDRAW EVOLUTION MODEL CHANGE PRONE
- 51. APACHE-ANT DECEMBER FEBRAURY ENTIRE RESTRUCTURING OF THE SYSTEM! 2006
- 52. APACHE-ANT DECEMBER FEBRAURY ENTIRE DESTRUCTING OF THE SYSTEM! ELIMINATION OF OLD CLASSES REDISTRIBUSTION OF RESPONSIBILITIES 2006
- 53. APACHE-ANT DECEMBER FEBRAURY ENTIRE DESTRUCTING OF THE SYSTEM! EVOLUTION MODEL 2006
- 55. Developer-based models work better DBCM +8%F-MEASURE than existing models
- 56. DBCM +8%F-MEASURE Developer-based models work better than existing models Highly complementarybetween models
- 57. Combination of these factors Relation between “changy” and buggy classes
- 58. Developer-Related Factors in Change Prediction:An Empirical Assessment May 22 and 23 Buenos Aires - Argentina Gemma Catolino*, Fabio Palomba°, Andrea De Lucia*, Filomena Ferrucci*, Andy Zaidman° *University of Salerno, °Delft University of Technology