ON THE USE OF GENETIC PROGRAMMING Adam C. Jensen and Betty H.C. - - PDF document

2/27/2014 1

ON THE USE OF GENETIC PROGRAMMING FOR AUTOMATED REFACTORING AND THE INTRODUCTION OF DESIGN PATTERNS

Adam C. Jensen and Betty H.C.

• Cheng. 2010. In Proceedings of

the 12th annual conference on Genetic and evolutionary computation (GECCO '10

Motivation

• Metrics enable a software engineer

to evaluate various characteristics

• f design quality
• Design patterns provide

contextual-driven solution template for solving design problems that occur frequently in large software projects

• Reduce the cognitive labor for

software engineers by suggesting design changes that are tailored to the software design in question

Presented by Rodrigo Morales

2/27/2014 2

Objective

• Automates the use of software engineering metrics to generate

refactoring strategies that introduce design patterns

• Using a an evolutionary computation-based approach REMODEL

Presented by Rodrigo Morales

Components

• Solution representation Refactored software design
• Mechanism for making changes to that solution
• Means of measuring the solution’s quality

Presented by Rodrigo Morales

2/27/2014 3

Solution Representation

• Refactored software design
• A design graph that represents the software design that is being refactored
• The set of steps to transform the original design into the refactored

design

• A transformation tree whose nodes represents minitransformations1

Presented by Rodrigo Morales

Metrics

• QMOOD[2] metric suite
• 11 individual metrics
• Evaluates distinct characteristics of OO design quality

Presented by Rodrigo Morales

2/27/2014 4

What is REMODEL?

• Genetic programming environment
• Guided by Software metrics
• To determine the optimal set of refactoring's to apply to a software

design

Presented by Rodrigo Morales

It looks for

• Improve the quality of the design as measured by the QMOOD
• Introduce design patterns when appropriate

Presented by Rodrigo Morales

2/27/2014 5

How individuals are represented?

• The individuals are represented as a pair that includes
• A design graph
• Transformation tree

Presented by Rodrigo Morales

Design graph

• Represents the software design that is being refactored

Presented by Rodrigo Morales

2/27/2014 6

Transformation tree

• Set of changes to the individual’s design graph
• Executing the tree involves visiting each of its nodes and executing

that node’s children recursively from left to right

• Each node is
• Transformation node (implementation of minitransformations)
• Information node
• Root node that

Presented by Rodrigo Morales Presented by Rodrigo Morales

2/27/2014 7

REMODEL Benefits

• The use of graph to represent target software design enables efficient

computation of metrics

• The order of the nodes describe a step-by-step series of refactoring

Presented by Rodrigo Morales

Design change mechanisms

• Abstraction
• Abstract Access
• Delegation
• Encapsulate construction
• Partial abstraction
• Wrapper

Presented by Rodrigo Morales

2/27/2014 8

Mutation Operations

• To explore the space of refactoring strategies:
• After being selected 90% of individuals undergo
• Subtree CrossOver
• Point mutation

Presented by Rodrigo Morales Presented by Rodrigo Morales

2/27/2014 9

Design Evaluation

• Prolog-based design pattern detector

Presented by Rodrigo Morales

Abstract Factory

• Abstract Factory
• Concrete Factory
• Abstract produt
• Conret Product
• Client

Optimizations

• 1. Encourages individuals to evolve transformation trees with fewer

transformation nodes

• 2. Rewards individuals that incorporate sequences of transformations

nodes produce design pattern fragments

Presented by Rodrigo Morales

2/27/2014 10

Fitness Function

Presented by Rodrigo Morales

Where: Q is quality PR: amount of fitness bonus of at least one design pattern NCP fitness penalty proportionate to the number of transformation nodes in Transformation tree MSR: fitness bonus given for the presence of specific sequences of transformation nodes C: coefficient that determines the relative weight of that term

Validation

• 4 empirical experiments (58 published software designs 5-15 classes

each)

• Case study ReMoDD Web-based software system
• Each experiment comprises 100 generations
• Tournament selection with a tournament size of 7
• REMODEL was applied 5 times

Presented by Rodrigo Morales

2/27/2014 11

Case Study: ReMoDD

• Repository for Model-Driven development containing model-driving

engineering artifacts (test cases, documentation, model and code)

• Comprises 23 classes

Presented by Rodrigo Morales

Discussion

Presented by Rodrigo Morales