Automated Software Quality Optimization: A MultiObjective Evolutionary Approach

1
Automated Software Quality Optimization A
Multi-Objective Evolutionary Approach

• Sepand Ansari
• Siavash Mir Arab

2
Importance

• Appearance of a new tool (late 50s)
• Knowledge based economy Mass production ? Mass
  Configurability
• Software Crisis (early 70s)
• The process of software development and
  reengineering is mostly manual because of its
  complexity.
• Software Requirements
• Functional
• Non-Functional ? Software Quality
• Maintainability (50 of software development
  budget)
• Reusability
• Complexity (increases debugging cost)

University of Tehran
3
Design Metrics

• Design metrics are used to quantify software
  quality
• Some renowned design metrics
• D
• CYC (Cyclic Dependency)
• EP (Encapsulation Principal)
• MQ (Modularity Quality)
• LCOM (Lack of Cohesion Metric)
• DIT (Depth in Tree)

University of Tehran
4
D Metrics

• Distance from main sequence

University of Tehran
5
Design Patterns

• First introduced by Christopher Alexander in
  Architecture
• a Design Pattern is a general repeatable solution
  to a commonly-occurring problem in software
  design.
• 23 Software Patterns introduced by Erich Gamma,
  Richard Helm, Ralph Johnson and John Vlissides
  (Gang of Four)
• Singleton pattern
• Factory pattern
• State pattern

University of Tehran
6
Design Patterns

• Meta-Patterns introduced by Batory at University
  of Texas and Tahvildari at University of Waterloo
• Low level patterns (in comparison with GOF)
• Building blocks of GOF patterns
• It is proved that software behavior is preserved
  after applying these patterns

University of Tehran
7
Problem Definition

• Problem Definition
• How to find the best sequence of design patterns
  which by applying them to the software design we
  achieve the best quality?
• Large search space (23 patterns with parameters
  that indicate where they should be applied)
• No complete and optimum algorithm
• discontinuity (invalid sequences)
• multiple optima

University of Tehran
8
Mapping to GA

• Chromosomes a sequence of supergenes each
  representing one pattern
• Crossover two crossover operators
• Inter-pattern crossover
• Intra-pattern crossover
• Mutation rate 0.06 Log (N) where N is number
  of classes
• Fitness Function D metric in single objective
  optimization

University of Tehran
9
Multi-Objective Evolution

• We want to improve different aspects of software
  quality (or at least preserve those that can not
  be ameliorated)
• Weighted sum of different objectives
• Difficult to define weights
• Vector Evaluated Genetic Algorithm (VEGA)
• For each objective, a natural selector selects
  individuals according the one objective

University of Tehran
10
ADE/ACE framework
University of Tehran
11
Results
University of Tehran
12
Results
University of Tehran
13
Publications

• M. Amoui, S. Mirarab, Sepand Ansari, C. Lucas A
  Genetic Algorithm Approach to Design Evolution   
  Using Design Pattern Transformation
  International Journal of Information Technology
  and Intelligent Computing and proceeding of
  International conference on Artificial
  Intelligence and Soft Computing 2006
• M. Amoui, Sepand Ansari, S. Mirarab, C. Lucas
  Automated Software Quality Optimization A
  Multi-Objective Evolutionary Approach submitted
  to IEEE International Conference on Software
  Maintenance 2006

University of Tehran
14

• Thank you!

University of Tehran