8th Korea-Russia International Symposium on Science and Technology - KORUS 2004

1
USING DESIGN PATTERNS FOR DESIGN OF SOFTWARE
ENVIRONMENT FOR RESEARCHES IN GENETIC ALGORITHMS

• Yuri R. Tsoy, Vladimir G. Spitsyn,
• Department of Computer Engineering
• Tomsk Polytechnic University
• neuroevolution_at_mail.ru

2
Report structure

• 1. Introduction
• 2. Software environment description
• 3. Genetic algorithm and design patterns
• 4. Flexibility and extensibility
• 5. Conclusion

3
Design patterns and software design

• Design pattern - the way of cooperation of
  objects and classes that are adapted for common
  design task solution in certain context. (by
  Erich Gamma, Richard Helm, Ralph Johnson and John
  Vlissides)
• Main advantages available with use of design
  patterns
• reusability
• flexibility
• extensibility

4
Researches in genetic algorithms

• Problems
• No specialized software environment for
  algorithms parameters settings and control,
  experimental data processing and visualization.
• No methodology of flexible software
  implementation of genetic algorithms.

5
Report structure
1. Introduction 2. Software environment
description 3. Genetic algorithm and design
patterns 4. Flexibility and extensibility 5.
Conclusion
6
Common program architecture
5 main blocks - Interface unit - Genetic
algorithm block - Task environment block - Data
analysis block - Visualization block
7
Interface unit

• Isolates all blocks from each other. All
  communications are made via Interface unit using
  unified data format.
• This helps to combine different implementations
  of different blocks.

8
Genetic algorithm block

• Provides implementation of some model of genetic
  algorithm.
• Block is separated from other systems blocks for
  purposes of use of different genetic algorithm
  models for one and the same task solution.

9
Task environment block

• Provides implementation of some task environment.
  Can be used for both modeling and evaluation
  purposes.
• Block is separated from other systems blocks for
  purposes of use of different tasks with one and
  the same genetic algorithm model.

10
Data analysis block

• Provides instruments and methods for analysis of
  data from both Genetic algorithm and Task
  environment blocks.
• Block is separated from other systems blocks for
  purposes of use of different data analysis
  approaches and implementations. In fact can
  consist of several sub-blocks.

11
Visualization block

• Provides visualization of data processed by Data
  analysis block or passed from Genetic algorithm
  and Task environment block.
• Block is separated from other systems blocks for
  purposes of use of different data visualization
  algorithms and methods.

12
General block structure
Control unit - provides routines for control of
blocks functioning. Processing unit -
implements functions specific for given
block. Internal data - stores internal block
data.
13
Report structure

• 1. Introduction
• 2. Software environment description
• 3. Genetic algorithm and design patterns
• 4. Flexibility and extensibility
• 5. Conclusion

14
Taking a deeper look at Genetic algorithm block

• Following parameters of genetic algorithm can
  vary depending on given GA model
• genetic representation
• population initialization
• population evaluation
• selection strategy
• reproduction strategy (parental pair selection)
• genetic operators
• next generation formation.

15
Taking a deeper look at Genetic algorithm block
(2)

• Parameters objects created by the Factory
  patterns
• genetic representation
• population initialization
• genetic operators
• Task (algorithm type) dependent parameters
  objects
• population evaluation
• Universal (task independent) parameters objects
• selection strategy
• reproduction strategy (parental pair selection)
• next generation formation.

16
Report structure

• 1. Introduction
• 2. Software environment description
• 3. Genetic algorithm and design patterns
• 4. Flexibility and extensibility
• 5. Conclusion

17
Flexibility and extensibility (1)

• Introduced architecture allows to
• Implement blocks as components of different
  specialized software packages.
• Combine different programs and make them work
  together using appropriate data and control
  adapters.
• Organize inner structure of certain block in any
  way.
• Combine blocks independently and transparently
  for other blocks. For example, Data analysis and
  Visualization blocks can be united in one larger
  unit without any changes in other blocks.

18
Flexibility and extensibility (2)

• Introduced architecture allows to
• Modify blocks without any changes in other parts
  of the introduced system. It is possible to
  change operators or encoding method in Genetic
  algorithm block but other blocks will never
  know about it.
• Extend blocks functionality independently. For
  example it is possible to add some new method of
  data analysis in appropriate block then make some
  changes in control adapter and use new ability.

19
Report structure

• 1. Introduction
• 2. Software environment description
• 3. Genetic algorithm and design patterns
• 4. Flexibility and extensibility
• 5. Conclusion

20
Conclusion

• The result of this project is intended for
  research purposes and will be distributed free.
  Introduced software architecture can be applied
  for many other research areas.
• First release approximately 2-nd quarter of
  2005. Source code and documentation will be
  available too.

21
Thank you for your attention!