Home

The Development of Agent Support for a Case

Contents

1. 52 REFERENCES James Odell Designing agents Using life as a metaphor Distributed Computing pages 51 56 July 1998 James Odell Objects and agents How do they differ Working paper v2 2 URL http jamesodell com publications html 1999 Tim O Shea The learnability of object oriented programming systems In OOP SLA 66 Proceedings pages 502 504 New York NY September 1986 ACM ACM Press Enric Plaza Josep Llu s Arcos and Francisco Mart n Inference and reflection in the object centered representation language noos Journal of Future Generation Computer Systems pages 73 188 1996 Enric Plaza Josep Llu s Arcos and Francisco Mart n Cooperative case based reasoning In G Weiss editor Distributed Artificial Intelligence meets Ma chine Learning Lecture Notes in Artificial Intelligence pages 180 201 Springer Verlag 1997 Enric Plaza Josep Llu s Arcos and Francisco Mart n Knowledge and expe rience reuse through communication among competent peer agents Interna tional Journal of Software Engineering and Knowledge Engineering 1999 M V Nagendra Prasad Distributed case based learning Article written for An dersen Consulting Center for Strategic Technology Research Thought Leader ship 1998 http www ac com services cstar cstrdcb13 html M V Nagendra Prasad Victor R Lesser and Susan E Lander Retrieval and reasoning in distributed case bases Journal of Visual Communication a
2. 59 60 CHAPTER 7 CONCLUSIONS AND FURTHER DEVELOPMENT repositories and network agents is of course a major difference 7 3 3 CBR Agents Most general CBR agents research focus on communication and cooperation in dis tributed case bases 51 49 53 48 50 47 56 6 52 34 32 42 Although ASCRARAD is not a networked system facilitating distributed case bases the architectural solution communication and collaboration techniques could be extended to meet requirements for distributed CBR Prasad et al 51 53 uses facilitator agents to gather a group of competent agents based on the specification of the problem The solution differs from ASCRARAD in that the facilitator doesn t know the capabilities of the individ ual agents prior to the gathering In ASCRARAD the task is to find which coordinator has the workforce group suitable for solving the problem Huang 28 proposes that the stored cases could play a more active role in the retrieval process by letting each case be represented as an active agent ASCRARAD s cases are all agents capable of autonomous activity and communication S nchez Marr et al 57 describes a similar approach to the ASCRARAD hierarchical structure of the case repository During retrieval meta cases containing discriminating attributes are queried to select which groups of cases should be candidates for reuse 60 Appendix A Usage user manual A 1 Specifications The tool is build u
3. 8 2 Divide the number of occurrences of a particular type in a single class by the total number of references to types in the class java applet Applet has 55 different types FORMULA 3 3 normalized count c N number of references to types in K C K Ti T K Ti ZUM c K Ti ii EXAMPLE 3 4 N 55 c Applet URL 3 Sum up all the results from 2 for all classes in a package and receive a sum of all normalized count c 7 s java applet contains 4 classes 21 22 CHAPTER 3 THE ENVIRONMENT AND AGENTS FORMULA 3 5 package P package count pc M number of classes in P pc P Ti 53 e j l EXAMPLE 3 6 package java applet M 4 4 pe java applet URL V c URL 0 524 j l 4 Divide the sum of all normalized s pc P 7 by the sum of the results from 3 for all packages in the repository FORMULA 3 7 normalized packagecount pc Pst PEU 5 EXAMPLE 3 8 0 524 pc java applet URL a m 0 131 22 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 23 5 Divide the sum of all normalized pc s pc P 7 by the total number of packages in the repository JDK1 1 8 contains 78 packages FORMULA 3 9 normalized pc pc Q number of packages P in the repository EXAMPLE 3 10 Sum of all pc for the type URL 1 790 Q 78 6 Calculate the significance of a type in a package by the following formula FORMULA 3 11 significance sign pc P Ti
4. Suns JDK1 X Known bugs None 68 A 2 MENUS 69 A 2 5 The Agent menu Start General usage Starts the Manager Agent which in turn starts other agents There are three kinds of agents Manager Agent Coordinator Agents and Workforce Agents Known bugs The Manager Agent is unable to compile classes contained in pack ages You can avoid this problem in some occasions by commenting out the package statement Stop General usage Stops the Manager Agent The Manager Agent itself controls all other starting and stopping of agents Known bugs If you stop the agent while it s working some agents may survive the killing of all invoked by the Manager Settings General usage Opens the Agent Settings Dialog You can specify how often the Manager Agent should invoke the reuse cycle in minutes Alternatively you can spec ify that the agent should start when a certain number of code lines has been reached You can also specify which type of retrieval the Agent should use You can choose between Reuse using Workforce which means that you try to match the target with all cases in the repository or Reuse using Organization which means that you use package matching to decide which cases the target should be matched against You can specify number of cases to be retrieved Lastly you specify the threshold This value tells the agents how well the cases has to match the target Known bugs None 69 70 APPENDIX A USAGE US
5. else die Unfortunately these heuristics does not give the wanted results alone Some packages that contain relevant classes doesn t always get selected I have therefore provided the Coordinator agents with some more knowledge Packages where the contained classes don t belong together firmly based on their relatedness and where few special types exist e g java lang and java util will hardly ever be selected I have therefore hard coded that these very general pack ages should always be selected The java lang package contains the classes that are 25 26 CHAPTER 3 THE ENVIRONMENT AND AGENTS most central to the Java language and Java depends directly on several of the classes mostly data types in the java util package 20 I believe this is a reasonable solution since these packages contain core classes used in almost any Java program I have also added two checks for class name similarity The similarities are calculated based on a string alignment algorithm where if a good alignment is found similarity is set to the length of the longest identical substring relative to the length of the whole alignment If the name of the target class is similar to the name of a package e g MyApplet s similarity to java applet is 0 43 the package gets selected If the name of the target class is similar to any name of a class contained in a package e g MyApplet s similarity to JApplet from javax swing is 0 75
6. pe r sign P T Ber EXAMPLE 3 12 0 131 0 023 sign java applet URL LES t 23 24 CHAPTER 3 THE ENVIRONMENT AND AGENTS The significance value stign is used in a Coordinators matching of a target case with a package case In other words the Coordinator estimates a relevance value for the target case s types based on the significance of types in the classes contained in a package This process assists in finding out if it is reasonable to believe that packages may contain case candidates for reuse Estimating the Relevance of a Target Type The estimation of a target case types relevance in a package is calculated in the fol lowing steps 1 Count the number of occurrences of a particular type in the target case FORMULA 3 13 target case p 2 Calculate the relevance of the type by the following formula FORMULA 3 14 relevance rel N rel q V clp 74 sign P 71 w 1 24 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 25 3 The total similarity is the sum of the selected relevances FORMULA 3 15 similarity sim N sim P pi S rel P gi i 1 4 If the total similarity sim P p is a positive number larger than 0 it s a match If the total similarity is larger than 0 5 the agent continues to live in memory If else the agent dies FORMULA 3 16 If sim P vi gt 0 then it s a match else if sim P yi gt 0 5 then it s a partial match
7. Replace with field or both Find Next button highlights the next matched string from the current cursor position It will automatically pass the end 65 66 APPENDIX A USAGE USER MANUAL of the file and go to the beginning of the file If there is no matched string a warning message will appear After a matched string has been found the Replace button is enabled Replace button replaces the matched string with the replacing string Re place all replaces all the occurrences of the searching string with the replacing string If there is no matched string a warning message will appear The radio buttons Up and Down specify the search direction If Match whole word only is checked the words instead of character streams are compared with the search String If Match case is checked the search is case sensitive Known bugs 1 If you click Find Next continuously some of the clicks will be ignored 2 In Search Dialog box when you do a search and find the string the replace button is enabled Before you replace the string if you either delete or insert some characters then replace the string the result are not what you expected So don t do anything else between find and replace Delete General usage If there is selected text highlighted in the text editor it is deleted Otherwise nothing happens Known bugs None Cut General usage If there is selected text highlighted in the text editor the selecte
8. organizational model It consists of a single agent A1 controlling the inter action with the user the interpretation of the user s actions and work the construction of a software component and the retrieval of cases The agent is dependent of a global retrieval mechanism similar to the one proposed by Tessem et al 65 The advantage with this architecture is that the Agent is capable of controlling the retrieval process in the background without user intervention due to its autonomous capabilities ay C1 C2 C3 CA C5 Cn Figure 4 1 Single Agent Architecture 4 1 2 Manager Workforce Architecture implemented A single agent may be pretty smart but the value gained from agents coordinating their actions by working in cooperation is greater than that gained from any individual agent This is where agents really come into their element 35 36 CHAPTER 4 AGENT ORGANIZATION The second approach Figure 4 2 is a multi agent system involving relatively uniform agents Workforce agents Awl to Awn with a Manager agent Am as facilitator which handles the interaction with the user and controls the retrieval process The Workforce agents have interaction capabilities for communicating with other agents local decision making capabilities for controlling the activities of the agent and the communications with the Manager agent and task level capabilities for perf
9. reasoner would generalize the rules automatically through the process of applying them to new situations Learning After reaching a search based solution to a problem a system can save that solution so that next time a similar situation occurs search would not be necessary It can also be important to retain in the case base information about the success or failure of previous solution attempts thus CBR offers a powerful model of learning This description of CBR is based on Aamodt and Plazas ideas as they are expressed in the article Case based Reasoning Foundational issues methodological variations and system approaches 1 2 5 Why Java Java is a class based object oriented programming language supporting encapsulation inheritance and polymorphism These features guide system organization and are in tended to promote the reuseability of the resulting software components Java is intended to allow application developers to write a program once and then be able to run it anywhere 26 This and that the Java naming scheme facilitates the 10 2 5 WHY JAVA 11 sharing of object classes among distributed developers without having to duplicate software modules contributes to reuseability Software development with Java can therefore be viewed as a constructive activity where system functionality is composed from a set of existing components Java Reflection enables Java code to discover information about th
10. retrieving a candidate case the agent provides the user with feedback about what it has carried out The agent gives information about which alternatives the program mer has The alternatives consist in either proposing adaptation of the retrieved case s with help from the reuse assistant to continue the adaptation independently from the assistant or to continue the programming with new searches for others and maybe more appropriate cases for potential reuse It is of course also possible to stop the agent put it to sleep for a while or to change other settings The programmer is com pletely free to follow the agents advice or to ignore it 29 30 CHAPTER 3 THE ENVIRONMENT AND AGENTS 3 3 The Editor bruker stein ascrarad testing Foo java File Edit Make Repository Agent Options class Foo nreo public FooCint b public int add int b publie static void main lt StringL args public static void main String args Figure 3 3 Agent supported Rapid Prototyping Environment The editor s graphical display is separated into two views see Figure 3 3 The edit window shown on the left allows the developer to build and edit a class The devel oper may either open a previously saved Java file or use the default class The class window shown on the right reflects the class being developed in the edit window The class window contains extracted signatures from the class being develope
11. select the most similar entry in the base class description and match it to this entry As the entries in the base class are selected mark them not selectable 27 28 CHAPTER 3 THE ENVIRONMENT AND AGENTS 3 The total similarity is the sum of the similarities of the selected matches in the target case All similarities are represented as a number in the interval 0 1 where 1 indicates maximal similarity Similarity between each pair of entries methods constructors fields are computed using string identity for types and string similarity for names For methods the simi larity is computed from three parts 1 Similarity in return type Identity gives similarity 1 otherwise 0 2 Similarity in method name Identity gives similarity 1 substring containment gives a similarity which is the size of the contained string relative to the con taining string It also runs a string alignment algorithm on the names and if a good alignment is found similarity is set to the length of the longest identical substring relative to the length of the whole alignment 3 Similarity in arguments If we let the collections of arguments for the two meth ods be represented as two bags A and B bags are like sets but may have multi ple occurrences of same element then the argument similarity can be given by the formula FORMULA 3 17 A B B A sim A B 1 Al IB The three similarities are weighted and add
12. ATION 5 2 Build a system exhibiting this feature of intelligence 3 Examine the behavior of the system in different task environments and with different initial conditions I decided to build an agent architecture performing intelligent actions in a complex environment where experimentation with different initial conditions and different task environments is possible 2 20 Agent Orientation vs Object Orientation Objects and agents share many things in common but in my view differences exist which makes a useful and important perspective for system development In traditional object orientation objects are considered passive because their methods are invoked only when some external entity sends them a message Software agents have their own thread of control localizing not only code and state but their invocation as well Such agents can also have individual rules and goals making them appear like active objects with initiative or as Jeffrey Bradshaw puts it objects with an attitude 8 In other words when and how an agent acts is determined by the agent Agents are regarded as autonomous entities because they can watch out for their own set of internal responsibilities Furthermore agents are interactive entities that are ca pable of using rich forms of messages These messages can support method invocation as well as informing the agents of particular events asking something of the agent or receiving a response to an
13. ER MANUAL A 2 6 The Options menu Font General usage Font brings a dialog box on the screen where you can change the font of the text Editor Default button restores the options defined by the author Known bugs If the option file can t be read or written then the changes don t show In this case the default options will be used Foreground General usage Foreground brings a dialog box on the screen where you can change the foreground color of the text Editor Default button restores the options defined by the program Known bugs If the option file can t be read or written then the changes won t show In this case the default options will be used Background General usage Background brings a dialog box on the screen where you can change the background color of the text Editor Default button restores the options defined by the program Known bugs If the option file can t be read or written then the changes won t show In this case the default options will be used A 2 7 The Help menu Help describes the function of every Menu item 70 A 2 MENUS 71 A 2 8 The CASE FOUND menu Only appears if a reusable case is found Start reasoner General usage Starts the reuse tool The reuse tool assists in matching signature features from the retrieved base cases to the class under development target case The user may accept or discard the suggestions In addition the reuse tool suggests how to reu
14. ESULTS SUMMARY 51 3 javax swing JEditorPane 4 javax swing JTextPane 5 javax swing text html HTMLDocument 6 javax swing text html StyleSheet 7 javax swing JDialog 8 javax swing JToggleButton It took 2 min and 20 sec to run on the repository built from version 1 3 of the Java M Standard Development Kit The repository consists of 3256 cases and 79 packages 6 3 Test Results Summary The test done by Tessem et al gave the following results It took about 2 minutes on a Sun SPARC 10 on a repository consisting of over 1700 cases The results are hardly comparable for the Java 2 tests since the size of the repositories differ and since the versions of the JDK s are not the same Tessem et al used version 1 1 4 of the JDK Even though the retrieval seems to give the wanted results it takes some time using the Java 2 based repository The time issue is however no longer as critical using agents The agents do their work in the background without interfering with the users work The results show that the Manager Coordinators Workforce architecture is faster than the Manager Workforce architecture Most of the time they also retrieve the same cases but it would be reasonable to conclude that the Manager Workforce Architec ture is more secure no cases are left un matched Some adjustments to the package matching heuristics should be considered see Section 7 2 3 CBR Improvements To create the complete class repository ta
15. R 3 THE ENVIRONMENT AND AGENTS as cases From these signatures one may also extract some knowledge about what kind of component this is For instance whether it is a collection of some kind The reuse component may suggest mappings between signatures of a retrieved case and the target and the user may accept or discard the suggestions In addition the reuse tool suggests how to reuse a class either by extension or by lexical reuse of source code if it is available Java s reflective capabilities are used to extract case descriptions from compiled Java classes and case based reasoning is applied to support retrieval and adaptation of reusable components The purpose of the tool is to localize poten tially reusable code and to support the programmer in her program understanding and adaptation of the code Research done by Tessem et al has shown that the set of features that can be automat ically extracted utilizing the Java reflective capabilities e g method signatures field types inheritance information etc can be effectively used to retrieve components for subsequent reuse The fact that case descriptions are extracted automatically increases the chances for acceptance of the approach in real programming environments and cir cumvents the high start up costs traditionally associated with establishing repositories of reusable components Figure 3 1 is a conceptual diagram of the Rapid Application Development Environ ment fro
16. R 3 THE ENVIRONMENT AND AGENTS significance of a type is hence a value used in the matching with the target case s types If a type in the target case matches a highly significant type in a package it increases the Coordinators willingness to fire an event to its workforce If it finds that classes in the package contain highly significant types for this particular target case it will pass the target case to the workforce for further matching Coordinators that have good match but not the best will keep on living in memory but they won t fire events See Estimating The Relevance of a Target Type There s reason to believe that these Coordinators may be relevant when the user continues with extending her class Therefore we keep them in memory for the next round of matching If the Coordinators workforce is found to not contain many significant types for the target case it dies Estimating the Significance of a Type The estimation of a types significance in a package is calculated in the following steps The package java applet containing four classes the class java applet Applet and the type URL is used as an example in a JDK 1 1 8 repository containing 78 pack ages Count the number of occurrences of a particular type in a single class FORMULA 3 1 class K count c type T 20 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 21 EXAMPLE 3 2 class java applet Applet Applet forshort type URL c Applet URL
17. The Road to ASCRARAD The Development of Agent Support for a Case based Reuse Application for Rapid Application Development Stein Inge Morisbak June 22 2000 ii Contents 1 Introduction 2 Theory and Method 3 The Environment and Agents 4 Agent Organization 5 Agent Interaction 6 Comparison and Test Results 7 Conclusions and Further Development A Usage user manual iz Installing and running the application References ill 15 33 39 47 53 61 73 77 iv Contents Chapter 1 Introduction To support programming in Java a tool Case based support for Rapid Application Development 65 has been developed which makes it possible to search through po tentially reusable Java classes The tool is based on case based methods to support retrieval and reuse The purpose of this project ASCRARAD is to add agent support to the retrieval mech anism of this tool This should optimize retrieval efficiency and free the user s hands and mind making it possible for the user to concentrate fully on her task Rapid Application Development is a technique used to obtain insight into system re quirements by experimenting with partial implementations To be effective it must be possible to create a prototype with a minimal investment of time and resources The purpose of this approach is to develop a prototype of portions of the proposed system to determine the necessity desirability or feasib
18. are devel opment knowledge In Proceedings of the 12th Annual Conference on Automated Software Engineering pages 46 53 IEEE Computer Society Press 1997 Ye Huang An evolutionary agent model of case based classification In Ian Smith and Boi Faltings editors Advances in case based reasoning third Euro pean workshop EWCBR 96 Lecture Notes in Artificial Intelligence Springer Verlag 1996 Michael N Huhns and Munindar P Singh Agents and multi agent systems Themes approaches and challenges In Michael N Huhns and Munindar P Singh editors Readings in Agents pages 1 23 Morgan Kaufmann 1998 Ivar Jacobsen Martin Griss and Patrik Jonsson Software Reuse Architecture Process and Organization for Business Success ACM Press New York 1997 P Katalagarianos and Y Vassiliou On the reuse of software A case based ap proach employing a repository In Automated Software Engineering volume 2 pages 55 86 Kluwer Academic Publisher Dordrecht Netherlands 1995 Craig A Knoblock Yigal Arens and Chun Nan Hsu Cooperating agents for information retrieval In Proceedings of the Second International Conference on Cooperative Information Systems UoToronto Press Toronto 1994 80 REFERENCES 81 33 Y Labrou and T A Finin A proposal for a new kqml specification tr cs 97 03 Technical report Computer Science and Electrical Engineering Department University of Maryland Baltimore County Baltimore Februar
19. be my chosen implementation strategy and to justify this choice Chapter 3 and 4 Chapter 5 is about agent language and interaction Some example runs will be presented in Chapter 6 I will conclude with a summary and some thoughts about further development and possibilities in Chapter 7 Chapter 2 Theory and Method The main purpose of this project is to implement an agent architecture suitable for retrieval of cases in a Case based Reasoning tool for Rapid Application Development I have decided to create an architecture where each case in the repository is an active case or agent and where a hierarchical structure of agents provide an organization analogy useful to implement the retrieval mechanisms and rules The big question was can agent support prove to be useful for retrieval of cases in a Rapid Application Development environment based on CBR techniques 2 1 AI Method Scientific method in artificial intelligence AI is often characterized by an experimen tal or explorative angle of incidence I have chosen to follow an approach based on 3 4 CHAPTER 2 THEORY AND METHOD this method Alis the study of the mechanisms underlying intelligent behavior through the construction and evaluation of artifacts that enact those mech anisms 36 p 755 This definition of artificial intelligence AI is a description of a method for experi menting rather than a theory about those mechanisms underlying intelligence It te
20. ble solution to this problem since ASCRARAD is a stand alone application with a well defined task specification Java is however more portable than other languages can move between heterogeneous hardware systems supports encapsulation inheritance and polymor phism and has high reuseability see Section 2 5 Extending and or adapting the environment may therefore not be very difficult tasks A common format for the content of communication is provided by the quasi KOML modeling of the ASCRARAD Agent Messages described in Section 5 0 5 Ontologies Ontologies are defined as specification schemes for describing concepts and their re lationships in a domain of discourse 19 It is important that agents not only have ontologies to conceptualize a domain but also that they have ontologies with similar constructions Such ontologies when they exist are called common ontologies Once interacting agents have committed to a common ontology it is expected that they will use this ontology to interpret communication interactions thereby leading to mutual understanding and ultimately to predictable behaviors The shared ontology in ASCRARAD is easily understood by the agents since we know that all agent interactions are concerned with retrieving base cases Java classes that 44 45 match a target case where a target case is a Java class developed by the user of the tool The vocabulary used in this process is small and the mea
21. brary and consists of 154 package cases and 5370 class cases The size of the repository is 18 MB I have created a smaller 1 2 2 repository by removing all sun packages and all com packages Only packages that are part of the Java M 2 Platform Standard Edition v 1 2 2 API Specification 61 are included in the repository Other packages may be a part of Sun s Java 2 SDK and Java 2 Runtime Environment distri butions The final repository consists of 2827 cases and 63 packages The size of the repository is 8 MB IBMs Jikes Compiler v 1 10 is used for compiling target case classes http ibm com developerworks opensource This document is created using IATEX 63 63 64 APPENDIX A USAGE USER MANUAL A 2 Menus A 2 1 The File Menu Open File General usage Open File brings a open file dialog box onto the screen in the dialog box you can traverse through the file system and pick the file to be opened into the editor If there is already a file in the editor the file will be replaced If the existing file in the editor has not been saved you will be reminded to do so Known bugs When the file is very big the text editor will crash Don t try to load files more than 1 000 lines If you accidentally load a big file it sometimes gets truncated In this case don t save the file Quit immediately Insert File General usage Insert File brings a file dialog box onto the screen in the dialog box you can trave
22. ce all operations hap pen in the search operation After the text is replaced the undo operation is able to undo the replacement Known bugs None 67 68 APPENDIX A USAGE USER MANUAL Redo General usage Redo reverses the undo operation Known bugs None A 2 3 The Make menu Compile General usage Compile brings a dialog box in which the users can enter options and arguments for the compiler For a description of these see the menus help file Known bugs When there are spaces in the file path directories it will try to convert them to a dos path For example d java source code test java will be converted to d javaso 1 test java since javac does not parse space in the file path However the conversion is not always correct and you have to correct it manually You can t have space in the source file name since javac won t recognize it A 2 4 The Repository menu Create Repository General usage Opens the Repository Maker Dialog In this dialog you can create a new repository or add cases to an existing The current CLASSPATH environment variable will by default appear in the CLASSPATH text field The Repository Maker uses this path to search for Java classes You may change this value by entering your own path to Java classes However it is not recommended to put untrusted classes into the repository This does NOT change the systems CLASSPATH variable Warn ing It takes long time to create a Repository of e g
23. collection of data that corresponds to some of the major slots of a KQML message 5 0 4 KQML KQML which is an acronym for Knowledge Query and Manipulation Language 33 was conceived both as a message format and a message handling protocol to support run time knowledge sharing among agents 19 This language can be thought of as consisting of three layers a communication layer which describes low level commu nication parameters such as sender and recipient a message layer which contains a 39 40 CHAPTER 5 AGENT INTERACTION performative and indicates the protocol of interpretation and a content layer which contains information pertaining to the performative submitted Keywords used in KQML messages are defined as follows 33 performative action such as requesting or commanding sender agent sending the message receiver agent receiving the message from original sender used when a message is sent using intermediary agents Not used in ASCRARAD to final recipient used when a message is sent using intermediary agents Not used in ASCRARAD in reply to identifier of the message that triggered this message submis sion Not used in ASCRARAD reply with identifier to be used by a message replying to this message Not used in ASCRARAD language language for interpreting the information in the content field of this message Not used in ASCRARAD ontology identi
24. d This view is used to compile and or interpret the current implementation When the developer invokes the agents the extracted signatures of the software is used as the basis for the target class As the developer extends changes or starts on a new class the Manager agent re reads the extracted signatures for each round of matching The Agent Settings Dialog is shown in Figure 3 4 In this dialog you specify how often the Manager Agent should invoke the reuse cycle in minutes Alternatively you can specify that the agent should start when a certain number of code lines has been exceeded When you ve reached the limit of e g 30 code lines the Manager will start 30 3 3 THE EDITOR 31 the retrieval process Next time it will start after reaching 30 more 60 code lines You can also specify which type of retrieval the Agent should use You can choose between Reuse using Workforce which means that you try to match the target with all cases in the repository or Reuse using Organization which means that you use package matching See Section 3 2 2 to decide which cases the target should be matched against Furthermore you may specify number of cases to be retrieved Lastly you specify the threshold This value tells the agents how well the cases has to match the target to be considered for retrieval and reuse CBR AgentNanager Settings EN a L Figure 3 4 Agent Settings Dialog A complete description of the menus
25. d text is deleted and stored in the buffer later it can be retrieved using paste Otherwise nothing happens Known bugs None 66 A 2 MENUS 67 Copy General usage If there is text selected highlighted in the text editor the selected text is copied into in the buffer later can be retrieved using paste Otherwise nothing happens Known bugs None Paste General usage If there is text in the buffer a cut operation or a copy operation has been done before the paste operation the text in the buffer is inserted into the text editor at the cursor position Otherwise nothing happens Known bugs None Undo General usage undo the operation just performed For example if a paste operation is just performed after the undo the pasted text disappears as if the paste operation never happened There is only one level of undo The operations that can be undone include insert delete cut paste replace and replace all For insert operation undo deletes the word you just typed A word is determined by two ways 1 undo deletes all the characters you typed until a space or a tab or a new line 2 after the cursor is moved by mouse or the arrow keys and some non space characters are typed undo deletes these characters For delete operation if you use backspace or delete key to delete a character undo reinserts the character If a chunk of text is deleted highlight and delete then the chunk of text reappears Replace and repla
26. d desires form its intentions plan what actions it will perform based on certain assumptions and eventually act on 7 8 CHAPTER 2 THEORY AND METHOD its plans 14 Furthermore intelligent agents must be able to interact with other agents using symbolic language Chapter 5 The interaction of many individual agents can give rise to secondary intelligent ef fects where groups of agents behave as a single entity An agent organization consists of individual agents acting according to their own rules and individual goals but the achievements as a whole can be viewed as greater than the sum of its individual con tributors Emergent intelligence is viewed as a phenomenon resident in and emerging from a society and not just a property of an individual Intelligence is reflected by the collec tive behaviors of large numbers of simple interacting agent So whether we take these agents to be neural cells individual members of species or single persons in a society their interactions produce intelligence ASCRARAD uses a metaphor from the organizational structure of an enterprise see chapter 4 The intelligence in the solution arise from the interactions of all the simple individual embodied agents All this sounds like models of rational human thinking and that is exactly what it is Our understanding of how humans think and how they act together in societies is used as a model for designing agents and agent systems o
27. e editor See Section 3 3 The Editor A menu pops up on the menu line containing to choices The user may choose to ignore the agents message by pressing Ignore or she may start the reuse tool The Manager also monitors the coding The user specifies how often a time interval or under what condition number of code lines in the editor the Manager should inter pret the code and make a target case for the retrieval process The Manager interprets the code following the user specification and passes a message to the next layer of agents namely the Coordinators 3 2 20 The Coordinator Agents Coordinators are package cases The logical grouping of cases and the concept of Coordinators are based on the Java concept Packages 26 Java programs are organized as sets of packages Each set has its own set of names for types which help to prevent name conflicts The naming structure for packages is hierarchical which is convenient for organizing related packages in a conventional manner A Coordinator agent represents a single package in the repository as a package case A package case consists of all the types method return types fields and argument lists of all the classes in a Java package Each value has a significance attribution The significance of a type in a certain package is a calculation of its number of occurrences in a case in relation to occurrences in each other package and in the whole repository The 19 20 CHAPTE
28. e look and feel and similar basic features These mecha nisms are especially helpful for newly employed and novices since they don t have the experience working with and the overview of the firm s repository This study and that of Tessem et al 65 has shown that the set of features that can be automatically extracted utilizing the Java reflective capabilities e g method sig natures field types inheritance information etc can be effectively used to retrieve appropriate components for subsequent reuse The introduction of agents in the environment has been successful They have proven to provide their retrieval tasks in time and more efficiently than if the user were con trolling the process The autonomous agents free the users hands and mind and makes it possible for the user to concentrate fully on her task The user may continue working while the agents provide helpful assistance in the background I believe I have reached my goals in developing a tool that fulfills the requirements I have set forth The tool helps the developer to locate code for potential reuse in an automated way and the tool aids in program understanding and adaptation It allows an exploratory approach to program development and optimizes reuse efficiency Can agent support prove to be useful for retrieval of cases in a Rapid Application Development environment based on CBR techniques I would say yes 7 2 Further Development I will in the f
29. e appears start the adaptation tool CASE FOUND Start Reasoner Check out the adaptation features of the adaptation tool 12 Appendix B Installing and running the application All files needed for the installation can be downloaded from http www ifi uib no staff stein ascrarad B 1 WinNT 2000 1 Create the directory C ascrarad 2 Copy ASCRARAD Zip to that directory 3 Unzip the ASCRARAD zip file inside the installation directory using WinZip can be obtained from www winzip com 4 Install Suns SDK 2 v 1 2 2 if you haven t already installed it can be obtained from http java sun com products jdk 1 2 13 74 APPENDIX B INSTALLING AND RUNNING THE APPLICATION 5 Include T7 c jdk1 2 2 lib dt jar c jdk1 2 2 lib tools jar c jdk1 2 2 jre lib il8n jar c jdk1 2 2 jre lib jaws jar c jdk1 2 2 jre lib rt Jjar c jdk1 2 2 jre lib plugprov jar in the systems CLASSPATH variable It is important that you include the complete paths to the jar files to make sure the jikes compiler works Include C jdk1 2 2 bin C ascrarad jikes in the systems PATH variable Double click the run_ascrarad bat file B 2 UNIX Solaris 1 Create the directory ascrarad in your home directory Copy ASCRARAD tar gz to that directory Unzip the ASCRARAD tar gz file inside the installation directory gzip d ASCRARAD tar gz Untar the ASCRARAD tar file inside the
30. e fields methods and constructors of loaded classes and to use reflected fields methods and constructors to operate on their underlying counterparts at runtime 26 This capability allows us to extract feature descriptions from compiled classes without having access to the source code Java has a set of powerful mechanisms that directly support software reuse However the developer must have a sufficient knowledge of the language environment to be able to construct a mental mapping from existing object classes to the class that he wishes to construct Java supports this to some extent in that it is possible to inherit the structure and functionality of an existing class and only specify new behavioral features in the new object class 2 5 1 Extending Java s Reuseability The software developer is left without support for acquiring an intimate knowledge of the contents of all the class libraries that are available for reuse This may explain why object oriented programming has such a steep learning curve 46 Much of the effort of becoming an effective programmer in an object oriented environment is expended in becoming familiar with the existing class libraries This is made even worse if the Java does not provide multiple inheritance The interface construct allows classes and their descendants to define and implement several interfaces as a set of methods An interface can then be used by other classes as a form of contract So wh
31. earlier query Lastly because agents are autonomous they can initiate interaction and respond to a message in anyway they choose In other words agents can be thought of as objects that can say No as well as Go 45 It is not entirely correct that OO doesn t support many of the properties ascribed to agents OO technology can be extended in various ways to support many of the prop 5 6 CHAPTER 2 THEORY AND METHOD erties ascribed to agents E g process and thread can be considered active objects The point here is that the agent based approach is an extension to how we think in an OO world Agents then are an evolution rather than a revolution It is another way of thinking about systems and their implementations i e an extension to the way of thinking in the OO approach The autonomous and interactive character of agents more closely resembles natural systems than do objects Since nature has long been very successful identifying anal ogous situations to be used in agent based systems is sensible We can e g think of agents as being a part of an environment where they have the potentiality of surviving and succeeding the ability to command resources and cooperate with others and the possibility of failure replacement and even death As an example the ASCRARAD solution uses an organization analogy The system is organized in departments where each department consists of a group of workers with special features that de
32. ed into a total similarity for the methods Similar approaches are used for constructors and data fields For constructors only argument similarity counts whereas for data fields type and name similarity counts 28 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 29 3 2 4 Back to the Manager The Manager collects all retrieved cases from the different packages The best cases are sorted by how well they match the target case and are presented to the user The leftover cases or workforce agents are kept alive as they may become potential cases for reuse in the further development of the target case In the next round of matching these leftover cases will be re matched without having to re read their features and invoke them again The manager agent s responsibility in this environment is to independently execute the case based matching cycle at the right time and with satisfying feedback The agent is able to monitor the users code and when the similarity to a case class is satisfying enough give feedback For this to work smoothly the agent has to know when to pass the problem to the right coordinator s when to disturb the programmers concentration and when to start retrieving a case proposed by the coordinators The standards for this should be based on the nature of the problem and the programmers skill level and experience It is possible for the users to adjust the system to their individual needs see Section 3 3 After
33. ed with the class for later use Each case file is associated with a Workforce agent When a base case is matched with a target case it obtains a value see Estimating the Similarity below This value between 0 1 determines if the case is a candidate for reuse If the match is good greater than a predefined threshold the Workforce agent offers it self as a potential case for retrieval The user specifies the threshold the case has to match to be considered as a potential case for reuse If the match evaluates to half of the threshold the Workforce agent continues to live in memory but does not send an event If not it kills itself There s also here as for the Coordinators reason to believe that these Workforce agents may be relevant when the user continues with extending her class Estimating the Similarity The estimation of the similarity between the target and the base is developed by Tessem et al and is described in the article Case based Support for Rapid Application Devel opment 65 The Workforce agents base cases estimate a similarity to the target class using simi larities between pairs of methods constructors and data fields To establish a similar ity for a base case it does the following steps 1 For each method constructor and data field in the base class use its signature to compute a similarity to each of the method signatures of the target 2 For each method constructor and data field in the target
34. es This re sults in an information owerflow between the agents and null values Workforce agents for instance don t need information about how many cases should be retrieved noCases The agentCase and content slots will have null values until it reaches a Work force agent The choice is made for simplicity flexibility and extendibility reasons A uniform message format is easier to maintain and extend Furthermore the meaning of the slots is easier to understand when they are common to all messages in the system The optional fields such as agent Case and content can be viewed as contain ers requesting to be filled according to the performative submitted I believe that my choice also makes it easier to extend the system to become networked see Section 7 2 Further Development A common format facilitating high level communication is important and essential for distributed agents on the internet 5 0 6 Interacting Agents Interaction is one of the most important features of the agents Researchers investigat ing agent communication languages mention three key elements to achieve multi agent 43 44 CHAPTER 5 AGENT INTERACTION interaction 19 29 A common agent communication language and protocol Acommon format for the content of communication A shared ontology The common agent communication language and protocol in ASCRARAD is hard coded using the Java programming language There is no need for a flexi
35. et a second chance in the next round of retrieval and who will die In stead of treating the cases in the case base as a passive mass of previous experience this model expands cases to become active agents with knowledge and intentions A uniform agent architecture is appealing A single computational framework is de veloped for one agent and then replicated as needed Each agent possesses the same social abilities as every other agent 4 1 3 Manager Coordinators Workforce Architecture implemented The last model consists of three agent layers where the agents are organized hierarchi cally using an enterprise analogy Figure 4 3 In the enterprise model the Workforce is split into departments p1 to pn led by Coordinator agents Acl to Acn On top of the organization is the Manger agent Am In this architecture the basic approach involves developing specialized agents whose organizational role is to handle the problem of finding the best men for the job This role has been realized in multiple facilitators called Coordinators The Coordi nators represent a layer of diversity in the organization providing a connectivity and routing layer in addition to the task level represented by the Workforce agents Co ordinators extract information about some basic features of the target case and decide whether the staff in her department has the qualifications for solving the problem see Section 3 2 2 The basic features extracted a
36. fies the ontology to interpret the information in the content field of this message Not used in ASCRARAD content context specific information describing the specifics of this mes sage 40 4 5 0 5 ASCRARAD Agent Messages I have added a couple of extra fields for the ASCRARAD environment agent messages These are noCases number of cases to be retrieved the user may specify how many cases she wants to retrieve targetCase the target case agentCase the base case threshold the user specifies the threshold the case has to match to be considered a potential case for reuse EXAMPLE 5 1 find case noCases 8 content null receiver coAgent i getName sender this getName targetCase MyString agentCase null threshold 0 75 A ASCRARAD Agent Message is shown above Example 5 1 The message starts with find case which is the action performative intended for the message There is no need for a protocol of interpretation language and ontology since ASCRARAD is a homogeneous environment with a single well defined language for interpreting the information in the content fields i e the agents know how to handle the content of the message based on the performative and the only language used in the environment 41 42 CHAPTER 5 AGENT INTERACTION is Java and text The remainder of the message contains keywords needed for the content and communication layers The co
37. ile Java only allows single inheritance for classes a single class can implement multiple interfaces and an interface can be defined as an extension of multiple other interfaces 30 11 12 CHAPTER 2 THEORY AND METHOD class repository is dynamic i e constantly being extended by multiple programmers The purpose of a case based retrieval and reuse tool is to help the developer to locate reusable code and to aid in program understanding and adaptation The tool matches Java classes from the class repository base cases to the target case the class under construction and then suggests similarities between them It is wrong to say that it extends Java s reuseability since the case descriptions are con structed using Javas own artifacts One could say though that it enhances reuseability in that it automates ensures the quality of and aids in understanding of the process Since the tool uses Java reflection and reuseability it doesn t require additional work on the part of the software developers Furthermore it ensures that the automated retrieval and adaption strategies will be immediately useful and work with existing software repositories However it should be noted that the same approach to feature extraction could be implemented in any other programming environment in which access to the source code for the components is available 2 5 2 Java Agents Java provides all of the functionality required to design and
38. ility of certain requirements The pur 1 2 CHAPTER 1 INTRODUCTION pose of such a prototype is to help the end user and developer to come to a mutual understanding of the system requirements and decide on a final design One mechanism for decreasing development time is the effective reuse of previously engineered components A software company may have a large repository of pre viously developed components and to make the reuse of these more effective will contribute a great deal to the development process These mechanisms are especially helpful for newly employed and novices since they don t have the experience working with and the overview of the firm s repository The ASCRARAD tool helps the developer to locate code for potential reuse in an automated way and in the background The tool aids in program understanding and adaptation It allows an exploratory approach to program development and optimizes reuse efficiency I will in this thesis present 7he Road to ASCRARAD In other words the theory behind and the process of the development of the agent support I will start with a the ory chapter describing the concepts of AI Method Agent Orientation Intelligence Case based Reasoning and the Java programming language Chapter 2 I will con tinue with a brief presentation of the tool developed by Bjgrnar Tessem et al before I go on to a more detailed description of the Agent support and architecture I will try to descri
39. implement software agents Besides Mobility a feature not used here it provides support for Autonomy and Arti ficial Intelligence Autonomy For a software program to be autonomous it has to be a separate process or thread Java applications are separate processes and may therefore last long and be autonomous An agent can be a single thread Java supports multithreaded applications and hence autonomy using both techniques The Java Virtual Machine offers a homogeneous interface for Java processes something that lets Java agents move between heterogeneous hardware systems 12 2 6 BRINGING IT ALL TOGETHER 13 Pattie Maes 37 head of MIT s Media Labs agents group defines an agent to be a process that lives in the world of computers and computer networks and can operate autonomously to fulfill one or more tasks 38 Agents are autonomous programs or processes These processes are always ready to respond to a user s action or a change in the environment The agents are informed about changes in the environment by messages see Chapter 5 being sent to them Intelligence Two main aspects of Artificial Intelligence AI are knowledge represen tation and algorithms manipulating these Knowledge representation is often based on the use of slots or attributes that store information about some entity and chains or references to other entities Java objects may be used to code this data and behavior as well as relations bet
40. installation directory tar xf ASCRARAD tar Install Suns SDK 2 v 1 2 2 if you haven t already installed it can be obtained from http java sun com products jdk 1 2 Include usr java lib dt jar usr java lib tools jar usr java jre lib exactvm Jjar usr java jre lib il8n jar 74 B 2 UNIX SOLARIS 75 usr java jre lib rt jar in your CLASSPATH variable It is important that you include the complete paths to the jar files to make sure the jikes compiler works 7 Include usr jdk1 2 2 bin S HOME ascrarad jikes in your PATH variable 8 Execute the run ascrarad sh file 75 76 APPENDIX B INSTALLING AND RUNNING THE APPLICATION 76 1 3 4 5 6 References Agnar Aamodt and Enric Plaza Case based reasoning Foundational issues methodological variations and system approaches AJ Communications 7 1 39 59 1994 K D Althoff A Birk Gresse von Wangenheim and C Tautz Case based rea soning for experimental software engineering In M Lenz B Bartch Sp rl H D Burkhard and S Wess editors Case Based Reasoning Technology From Foundations to Applications pages 235 254 Springer Verlag 1998 Klaus Dieter Althoff Markus Nick and Carsten Tautz Concepts for reuse in the experience factory and their implementation for cbr system development In Proceedings of the 1 1th German Workshop on Machine Learning August 1998 J R Ander
41. kes several hours for Java 2 libraries but 51 52 CHAPTER 6 COMPARISON AND TEST RESULTS since this is done only when the class library is updated I don t consider the efficiency of this to be of particular significance 52 Chapter 7 Conclusions and Further Development 7 1 Conclusions Rapid Application Development is a technique used to obtain insight into system re quirements by experimenting with partial implementations To be effective it must be possible to create a prototype with a minimal investment of time and resources The purpose of this approach is to develop a prototype of portions of the proposed system to determine the necessity desirability or feasibility of certain requirements The pur pose of such a prototype is to help the end user and developer to come to a mutual understanding of the system requirements and decide on a final design One mechanism for decreasing development time is the effective reuse of previously engineered components A software company may have a large repository of pre viously developed components and to make the reuse of these more effective will 53 54 CHAPTER 7 CONCLUSIONS AND FURTHER DEVELOPMENT contribute a great deal to the development process Not only will it be more effec tive it will also contribute to a development where a certain company style will be followed An example of this is look and feel Software coming from the same firm should have the sam
42. lls us that the mechanisms are best studied through design execution and evaluation of experiments with improvement of models and further experimentation as a goal As an empirical science AI takes a constructive approach we attempt to understand in telligence by building a working model of it Hence AI is also engineering AI uses much of the theory we find in cognitive psychology 43 54 4 16 evolu tionary psychology 5 11 12 and organizational and social psychology 40 21 35 We wish to let computers perform tasks which if humans carried them out would be viewed upon as something demanding intelligence The task for AI research is to find tasks demanding intelligence and to see what processes are sufficient to accom plish these Through experimentation with alternative models and implementations you can find out much about the principals for intelligence both individual and corpo rate Principles that are best understood through repetitively experimentation trying and failing The moment of truth is a running program 59 p 96 Herbert A Simon 59 states that the main method within AI consists of building and studying systems that exhibit intelligence Three points sum up the method 1 Choose a task incorporating a feature of intelligence that is of substantial practi cal importance or that exhibits features and complexities that have not yet been simulated by AI systems 2 2 AGENT ORIENTATION VS OBJECT ORIENT
43. lso be helpful for the Manager agent of the ASCRARAD The Manager could test run uncompiled Java statements and thereby check the syntax of the expressions in an automated way The goal is to create a full featured Rapid Application Development Environment 55 56 CHAPTER 7 CONCLUSIONS AND FURTHER DEVELOPMENT 7 23 5 CBR Improvements Possible CBR improvements are Improve the package matching It should be possible to improve the pack age matching so that fewer packages are selected during retrieval An example of how this could be done is to somehow find out what kind of class is being developed Is it e g a GUI class a collection of some sort or an IO class By extracting knowledge to recognize different types of classes either automatically using some heuristics or by letting the user specify it herself we should be able to select candidate packages better and to leave non candidates unselected Inferring features of a class to other kinds of features It should e g be possible to extract features from design patterns 24 Should also be possible to extend the case based approach to include cases containing reuse experiences and not only classes This could be done by indexing a case with a specification and perhaps a class reused and how it was reused Similarity can then be estimated based on similarity between the specifications in two cases To avoid the process of querying all coordinator agents each ti
44. m the article Case based Support for Rapid Application Development 65 An earlier version of the system included a Java interpreter This is shown in the figure as a grey box since it is no longer part of the environment As development progresses portions of the software could be sent to the interpreter for immediate feedback This is a great feature in RAD tools and should be introduced again See section 7 2 2 Editor Improvements 16 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 17 Extracted signatures Editor P 7 under deviciment pe x 4 Interpret TypeCheck ByteCompiled Mapper Class Repository Case Based Retriever Figure 3 1 Conceptual Diagram of Rapid Development Environment 3 2 Agent Supported Case based Retrieval A user of such a tool must repetitively and actively ask for reuse support An alter native to this is one or more intelligent agents observing the implementation the user builds and when the similarity to a class is good enough notifies the programmer The Agent Supported Case based Reasoning Application for Rapid Application Develop ment ASCRARAD tool is an attempt to support these features It is based on the work done by Tessem et al and extends their work with the implementation of agent support 17 18 CHAPTER 3 THE ENVIRONMENT AND AGENTS Figure 3 2 is a conceptual diagram of the Rapid Application Development Environ ment with Agent
45. me a new re trieval cycle starts and all workforce agents for selected packages it could be possible to rank the cases based on how often they are selected and how well experiences with the reuse of them are Good cases both coordina tors and workforce would then fire more often than cases we don t have very good reuse experiences with S nchez Marr et al 57 proposes an evolutionary solution where agents forget seldom used cases and prefer often used cases 56 7 3 RELATED RESEARCH 57 Convert the repository from files to some sort of database for efficiency Lastly and perhaps a bit far fetched at the moment it could be possible to extract case features from the comments in a class This would require advanced language interpretation 7 2 4 Evaluation An evaluation of the tool in a realistic environment is important to see the effects case based reuse support may have on target users These experiments should be executed with real users in a real environment where their tasks are similar to what they do every day The perfect user would be a newly employed software programmer working in a software company where rapid prototyping is an important task The company should have a repository of previously developed software components and a wish to improve reuse and efficiency in the software development process 7 3 Related Research There have been several attempts to use case based reasoning CBR t
46. nd Image Representation Special Issue on Digital Libraries 7 1 74 87 1995 M V Nagendra Prasad Victor R Lesser and Susan E Lander Learning orga nizational roles in a heterogeneous multi agent system In Proceedings of the International Conference on MultiAgent Systems 1996 Japan 82 REFERENCES 83 53 M V Nagendra Prasad and Enric Plaza Corporate memories as distributed case libraries In Procceding of the 10th Banff knowledge acquisition for knowledge based system workshop number 40 in 2 pages 1 19 Banff Canada 1995 54 Z W Pylyshyn Computation and Cognition Toward a Foundation for Cognitive Science MIT Press Cambridge 1984 55 Bergmann R and Eisenecker U Fallbasiertes schlieen zur untersttzung der wiederverwendung objektorientierter software Eine fallstudie In Proceedings der 3 Deutschen Expertensystemtagung XPS 95 1995 56 M Redmond Distributed cases for case based reasoning facilitating use of multiple cases In Proceedings AAAI 90 1990 57 M S nchez Marre J B jar and U Cort s Reflective reasoning in a cbr agent In Procc of the VIM Project Spring Workshop on Collaboration Between Human and Artificial Societies Spain 1997 58 Barry G Silverman Nabil Bedewi and Alfredo Morales Intelligent agents in software reuse repositories In Proc of ACM 4th InternationalConference on In formation and Knowledge Management CIKM 95 Baltimore Mar
47. nd reasoned about are data types used in the target case code Furthermore the Coordinators also check for string sim ilarity in package and class names If a target case doesn t contain common data 37 38 CHAPTER 4 AGENT ORGANIZATION amp 69 CP e DE C m Cm m CNET Ple Pao Ps pre eee eejee ee ee Figure 4 3 Manager Coordinators Workforce Architecture types for the type of cases in the department or the package doesn t contain classes with similar names to the target the Coordinator will avoid sending messages to its workforce Hence the Coordinator layer s task consists of forwarding messages to ap propriate agents Each department is constructed based on the package concept of Java see 3 2 2 and 26 The enterprise model provides greater efficiency than a completely uniform multi agent system since the invocation of all the task level agents is avoided However it recognizes only a single type of organizational diversity that of the coordinators In other words only the coordinators may decide if a problem is their department The task level agents have nothing to contribute with unless their coordinator sends them a message 38 Chapter 5 Agent Interaction The agents communicate by sending messages to each other All of the communica tions between the Coordinators Workforce and the Manager is done by sending Agent Messages These messages are basically a
48. ning of the words is easily understood An example of a situation where agents might misunderstand each other because of two slightly different ontologies is two English speaking agents getting confused if one talks about the boot and bonnet and the other about the hood and trunk of an automobile They have to have a shared vocabulary of words and their meaning The agents does not share the same ontology 45 46 46 CHAPTER 5 AGENT INTERACTION Chapter 6 Comparison and Test Results In the following I will present test results for both of the implemented architectures See Chapter 4 Only the test results under the WinNT environment will be presented see Appendix A 1 for specifications I have created two repositories for testing the tool The first is created from version 1 1 8 of the Java Development Kit 62 The repository consists of 1581 cases and 78 packages The second is created from version 1 3 of the Java M Standard Develop ment Kit The repository consists originally of 7174 cases and 241 pack ages I have created a smaller repository by removing all sun packages all com packages and the netscape javascript package Only packages that are part of the Java M 2 Platform Standard Edition v 1 3 API Speci fication 62 are included in the repository Other packages may be a part 47 48 CHAPTER 6 COMPARISON AND TEST RESULTS of Sun s Java 2 SDK and Java 2 Runtime Environmen
49. ntent of a message is a Workforce agent which in this case is nu11 since the message is a query from the Manager agent and has not yet reached a Work force agent The message does not contain a agentCase for the same reason A agentCase is the description of a retrieved base case from the repository and is added by the appropriate Workforce agent The last field belonging to the content layer is the targetCase field which is the case extracted from the class under de velopment The target case is in fact the content of the Manager agent a single base case is a Workforce Agent while the target case is a Manager The sender and receiver parameters specify information at the communica tion level sender is the name of the agent sending the message while receiver is the name of the agent the message is aimed at A Workforce agents receiver parameter is always the Manager The following is an example of a message from a Workforce agent EXAMPLE 5 2 case matched noCases 8 content this receiver Manager sender this getName targetCase MyString agentCase java lang String threshold 0 75 42 43 Example 5 3 shows a message from a Coordinator to a Workforce agent EXAMPLE 5 3 match noCases 8 content null receiver workAgent i getName sender this getName targetCase MyString agentCase null threshold 0 75 I have decided to maintain only one format for ASCRARAD agent messag
50. o represent the experience acquired in the design of past software components CBR is especially suitable when dealing with object oriented design where software development usually relies on previously developed object classes In the Java programming language the base case descriptions can be constructed from the software artifacts themselves using Java reflection Java is also suitable for imple menting agent societies since it supports multithreaded applications These choices of method techniques and models have been used as the basis for the development of ASCRARAD 14 Chapter 3 The Environment and Agents 3 1 Case based Support for RAD Rapid Application Development RAD is a technique for achieving insight into sys tem requirements by experimenting with partial implementations Reuse of previously developed components assists this process Tessem et al 65 have developed a tool where the reuse method is case based and integrated in an environment specialized for rapid prototyping The environment consists of an integrated editor and a reuse tool The reuse tool can be used on partial class specifications to identify components that may be candi dates for reuse In addition it supports the reuse process of the chosen components The case based reuse tool supports retrieval and reuse of classes based on their sig natures methods return types and arguments etc which in this case is viewed upon 15 16 CHAPTE
51. o support soft ware engineering but few combine the technique with agent support 7 3 1 CBR and Software Engineering Some approaches to support software reuse using CBR focus on the earlier phases of a software development process 41 39 60 64 others discuss the possibility of reuse of code by identifying similarity in designs or specifications assuming that this similarity leads to similarity in code 13 66 3 and a couple of authors focus on 57 58 CHAPTER 7 CONCLUSIONS AND FURTHER DEVELOPMENT case based reasoning as a tool to enhance organizational learning in software factories 27 2 Some case based tools to support the coding phase are Bhansali and Harandi s work on synthesis of UNIX scripts 7 Fouqu and Matwin s work on the case based reuse of C code blocks 22 Gomes and Bento s work on automatic conversion of procedu ral VHDL programs into cases 25 using extracted functional and behavioral knowl edge from basic language constructs Bergmann and Eisenecker s work on reuse of object oriented software in Smalltalk 80 and C 55 with weight on the definition and structuring of relevant program features used for indexing Katalagarianos and Vassiliou s work on reuse in an object oriented repository 31 taking into account only semantic similarity to the name of the component and Fern ndez Chamizo et al s work on retrieval and adaptation of object oriented classes 18 17 integrating code natu
52. of the graphical interface can be found in Ap pendix A 2 3l 32 CHAPTER 3 THE ENVIRONMENT AND AGENTS 32 Chapter 4 Agent Organization In order to get a large number of software agents to work together to perform complex activities effectively you have to choose some form of agent organization or architec ture Competent agents that share common communication skills viewpoints and purposes form an agent society An agent society is a very robust and flexible entity Agents within a society can interact with one another due to common language see Chap ter 5 identify the abilities and needs of each other due to common viewpoints and request or perform activities on behalf of others due to common goals Societal agents can determine that activities needs to be performed and enlist other agents to help out The flexibility of an agent society comes at a cost however a substantial inefficiency in handling ongoing or repetitive activities This cost begins with the prob lem of finding appropriate agents for an activity and continues through planning and 33 34 CHAPTER 4 AGENT ORGANIZATION executing the activity The flexibility of an agent society starting at square one incurs the cost of always starting at square one Agent societies provide a basis for developing more structured agent organizations What is an agent organization For Galbraith 23 an agent organization is 1 composed of competen
53. ollowing include some suggestions and thoughts concerning further de velopment of this tool There are many possibilities for extending and enhancing the 54 7 2 FURTHER DEVELOPMENT 55 solution Some a bit far fetched and requiring future research and some that could have been developed right away 7 2 1 Agent Improvements Distributed case libraries on the Internet is a challenging possibility for further devel opment of the agent support This can be made possible utilizing Javas capabilities of letting agents move between heterogeneous hardware systems KQML messages and Java reflection A lot of research is going on in the field of corporate memories and distributed case libraries 51 53 32 48 49 50 52 7 2 2 Editor Improvements I would like to improve the editor with more features On the Make menu I would like to add a Run method for running the application under development a View Applet for viewing applets and also a class browser Furthermore I would like to re introduce the Java interpreter from an earlier version of the Case based RAD tool developed by Tessem et al mentioned in the Case Based Support for Rapid Ap plication Development 65 article The Java interpreter allows the user to test run uncompiled Java statements thus the user can import Java classes to the environment and test their functionality in the interpreter without the need for compiling the appli cation This feature could a
54. omponents Expert Systems with Applications 9 3 397 405 1995 T Finin Y Labrou and J Mayfield Kqml as an agent communication lan guage In Jeffrey Bradshaw editor Software Agents pages 291 316 The AAAI Press The MIT Press 1997 David Flanagan Java in a Nutshell O Reilly Sebastopol CA 2 edition May 1997 ISBN 1 56592 262 X Ken Ford Clark Glymour and Patrick J Hayes Android Epistemology MIT Press Cambridge 1995 ISBN 0 262 06184 8 Gilles Fouqu and Stan Matwin A case based approach to software reuse Jour nal of Intelligent Information Systems 1 165 197 1993 Jay R Galbraith Organization Design Addison Wesley 1977 E Gamma R Helm R Johnson and J Vlissides Design Patterns Elements of Reusable Object Oriented Software Addison Wesley professional computing series Addison Wesley Publishing Company Reading MA 1995 79 80 25 26 27 28 29 30 31 32 REFERENCES Paulo Gomes and Carlos Bento Automatic conversion of vhdl programs into cases In Challenges for Case based Reasoning Proceedings of the ICCBR 99 Workshops University of Kaiserslautern Centre for Learning Systems and Ap plications 1999 James Gosling Bill Joy and Guy Steele The Java Language Specification The Java Series The Java Series Addison Wesley Reading MA 1 edition 1996 ISBN 0 201 63451 1 Scott Henninger Tools supporting the creation and evolution of softw
55. orming the work of the agent The difference between the Workforce agents in this architectural structure is only their content The content of each Workforce agent is an actual case description S A X a NC Va A a B N Aw1 Aw2 Aw3 Aw4 Aw5 A Awn a i F Figure 4 2 Manager Workforce Architecture Active Cases Traditionally cases are viewed upon as passive objects waiting to be retrieved for fur ther decision taking or problem solving The decision whether cases should be re trieved is taken by a global mechanism Ye Huang proposes in An Evolutionary Agent Model of Case Based Classification 28 to let the stored cases play a more active role The idea is to increase the flexibility of the retrieval process and to let each case reflect over its own context and to let cases work together to obtain global goals The active cases or Workforce agents are capable of matching themselves with a target case description received from the Manager Agent Based on the matching results they may decide to offer themselves as potential cases for reuse decide to wait for a further 36 4 1 CHOICE OF ORGANIZATION 37 development of the target case that may match better its description or simply die If an agent decides to offer its case description as a potential candidate for reuse it has to compete with other agents The Manager agent controls this war and decides who will be offered to the user who will g
56. r as James Odell says it we should use life as a metaphor 44 to build these systems Since the term intelligent is somewhat controversial in computer science and perhaps not applicable to the type of agents I have developed I will use agent not intelligent agent as the term for the type of agents in this system Instead I will assume a level of agent competency sufficient to allow them to communicate and work together to per form useful tasks to accept or infer instructions or requests regarding its activities and to use these to shape its autonomous activity decisions Such agents are the building blocks of the agent organization The agent system falls somewhere between a simple 2 4 CASE BASED REASONING 9 event triggered program and one with human collaborative abilities It s probably far to the former side of the spectrum 2 4 Case based Reasoning Case based reasoning CBR is a general method for reasoning on the background of experience CBR is a retainment model for representation indexing and organization of previous cases and a process model for retrieval and modification of old cases and assimilation of new ones A case usually describes a problem situation A previously experienced situation which has been captured and learned in a way that it can be reused in the solving of future problems is referred to as a base case past case previous case stored case or retained case A new case target case is the de
57. ral language documentation and expert domain knowledge under the same component representation scheme and of course the approach taken by Tessem et al 65 extracting features directly from the class definitions without the need for neither high level code or documentation 7 3 2 Agents and Software Reuse A multi agent system for searching and retrieving reusable software components has been developed by Erdur et al 15 Their work aims at designing and implementing a component retrieval module of a component based reuse environment as a multi agent system consisting of different types of software agents The author truthfully claims that It can be expected that in a very near future thousands of reusable component servers will be marketing their components on Internet and Internet will become the software repository of any organization 15 58 7 3 RELATED RESEARCH 59 The agents in this system are capable of retrieving components based on different ontologies Although the component ontology is based on the Basic Interoperability Model BIDM 9 there is no support for automated component creation or feature extraction from the software components themselves Hence the descriptions will have to be created manually Other ontologies used by the agents are domain ontology e g Type of component abstraction operations data types programming language and system ontology e g Server ip address domain type of components sto
58. red location The approach does not use Case based Reasoning for retrieval of components The searching is based on a traditional browse and surf mode facilitating agent capabili ties for search collaboration using KQML and handling of the dynamically changing nature of the Internet Silverman et al 58 explores the role for intelligent agents to support reuse in dis tributed software repositories The author has directed four separate repository design and evaluation projects where the architectures contain a variety of intelligent agents The most related types are the CBR ones These agents are not described in great detail but the general idea is similar to my approach The agents examine the user s description however incomplete of the target problem compares it with a repository of previous cases and retrieves the most similar case s Furthermore the author sug gests an adaptation facility Other learning agents such as design rationale capture learning by interviewing and analytical learning after tracking and studying outcomes are present in the system The author does not mention anything about how the reusable software artifact repos itories are created The last approach seems to be the most similar to ASCRARAD but since the specific details on the CBR agents and the repository creation is not explained in great detail itis hard to compare the two systems That ASCRARAD doesn t facilitate distributed
59. rse through the file system and pick the file to be inserted into the editor The contents of the file are inserted at the current cursor position Known bugs When the file is very big the text editor will crash Don t try to insert files more than 1 000 lines If you accidentally insert a big file it sometimes gets truncated In this case don t save the file Quit immediately New General usage New starts a blank text file The current file in the editor will be replaced If the existing file in the editor has not been saved you will be reminded to do so Known bugs None 64 A 2 MENUS 65 Save General usage Save saves the file in the editor If the file is new and does not have a name yet Save works the same as Save As Known bugs None Save as General usage Save As brings a Save As file dialog box onto the screen You can choose a directory from the directory list in the dialog box and give the file a name The file in the editor will be saved in the directory and the name you just specified If the file name you specified is already being used you will be asked whether you want to overwrite the file or not Known bugs None Quit General usage Quit terminates the program If there is a unsaved file in the editor you will be reminded to save Known bugs None A 2 2 The Edit menu Search General usage Search brings a Search dialog box on the screen you can fill either the Find field or the
60. s Extracted signatures Nw Version under development SE eee Interpret TypeCheck Coordinators ByteCompiled Workforce es 4 4 322 a amp y 8 EGE GEE Figure 3 2 Conceptual Diagram of Rapid Development Environment with Agents The Manager agent monitors the programmer s implementation via the extracted sig natures of the partial class specification and compiles the code when needed A re introduction of the interpreter would allow the Manager to test run uncompiled Java statements thus check the syntax of the expressions This is however not implemented but may be a great idea for further development See section 7 2 2 The Manager agent then passes the task on to the appropriate Coordinator agents Each Coordinator have more specific knowledge about a certain group of cases If a Coordinator thinks that his group of cases may be relevant for the user it sends a message to the workforce agents in this group Such a group is in reality the class members of a Java package 18 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 19 63 and a single workforce agent is in reality a single case description of a Java class 3 2 1 The Manager Agent On the top level we find the Manager agent The Manager handles the interaction with the user and controls the retrieval process When reusable cases are found the user is notified by a Case Found message in a small window at the bottom of the sourc
61. scription of a new problem to be solved The technique utilizes similarities in problem descriptions to select and adapt previous solutions to new situations Case based reasoning is therefore a cyclic and integrated process that consists of solving a problem learning from this experience solving a new problem etc Generally a CBR cycle may be described by the following four processes 1 RETRIEVAL of the most similar case base case 2 REUSE of the information and knowledge in the base case Adaptation and mapping of the base case to the present problem target case 3 REVISE the proposed solution evaluate and correct 4 RETAINMENT of the parts of this experience that may be useful for future problem solving The new case problem description and adapted solution is stored in the case base for future reuses the learning process 9 10 CHAPTER 2 THEORY AND METHOD Instead of general knowledge about a domain the system stores case descriptions and their associated solutions that are retrieved adapted and utilized in the solution of new problems There are many advantages with this approach Some of them are Simple knowledge acquisition We record a human expert s solutions or experiences from real life to a number of problems and let a case based reasoner select and reason from the appropriate case Generalization using cases This saves us the trouble of building general rules from the examples Instead the
62. se a class either by extension or by lexical reuse of source code if it is available In this part of the tool the similarities between signatures are used in the mapping process When the tool is asked to establish mappings it only establishes those with a similarity higher than some threshold 0 75 asks for verification of those not so similar 0 5 0 75 and discards the rest Depending on the total similarity between the cases the system suggests direct reuse subclassing extension or literal reuse copy and paste Known bugs None Ignore General usage Ignores the CASE FOUND message from the Manager Agent and removes the CASE FOUND menu Know bugs None 71 72 APPENDIX A USAGE USER MANUAL A 3 Scenario Open the MyApplet java file in the C ascrarad testing di rectory SHOME ascrarad testing on UNIX File Open Go to the Agent menu and start the Agent Settings dialog Agent Agent Settings Type in The number of cases you want to retrieve e g 5 The time interval for the Manager Agent i e how often it should start the retrieval processes e g 5 minutes Set the retrieval threshold e g 0 50 AVERAGE Choose Using Organization from the Action menu Click OK Go to the Agent menu and start the Agent Agent Start While the agent is working you may extend the class change it or open a new file When and if the CASE FOUND messag
63. sing Microsoft Visual J 6 0 and Emacs 20 6 1 with Java Developer Environment JDE 2 5 1 I started developing the product with MS Visual J It has great debugging facilities but doesn t support Java 2 have therefore ported the solution to Emacs See Appendix B for installation specifications Can be run under Suns Java Development Kit v1 1 8 or higher It is however recommended to use the latest JDK If you want to create repos itories from Development Kits higher than v1 1 8 you must include the paths to these in your CLASSPATH variable Has been tested on WinNT 4 0 with service pack 6 61 62 APPENDIX A USAGE USER MANUAL The WinNT box I ve used is a Pentium II 450MHz with 256MB RAM Two repositories are built and tested on WinNT They have the following features The first is created from Suns Java DK 1 1 8 class library and consists of 78 package cases and 1581 class cases The size of the repository is 3 84 MB The second is created from Suns Java 2 SDK 1 3 class library and consists of 241 package cases and 7174 class cases The size of the repository is 18 5 MB I have also created a smaller 1 3 repository for testing purposes see Chapter 6 for details 62 A l SPECIFICATIONS Has also been tested on Solaris7 The Solaris box is a Sun Enterprise 450 2x UltraSPARC II 400MHz with 1024MB RAM The repository on Solaris is created from Suns Java M 2 SDK 1 2 2 class li
64. son Arguments concerning representations for mental imagery Psy chological Review pages 249 277 1978 J H Barkow L Cosmides and J Tooby The Adapted Mind Oxford Univ Press New York 1992 R Barletta and W Mark Breaking cases into pieces In Proceedings of Case Based Reasoning Workshop St Paul MN 1998 TT 78 7 8 10 11 12 13 14 15 REFERENCES Sanjay Bhansali and Mehdi T Harandi Synthesis of UNIX programs using derivational analogy Machine Learning 10 7 55 1993 Jeffrey Bradshaw An introduction to software agents In Jeffrey Bradshaw editor Software Agents The AAAI Press The MIT Press 1997 Shirley Browne and James W Moore Reuse library interoperability and the world wide web In Proceedings of the 1997 international conference on Software engineering pages 684 691 Boston United States May 1997 Daniel D Corkill and Susan E Lander Organising software agents The impor tance of design to effective system performance Object Magazine 8 2 41 47 April 1998 Leda Cosmides and John Tooby The Adapted Mind Evolutionary Psychology and the Generation of Culture chapter Cognitive adaptations for social exchange Oxford University Press New York 1992 Jerome H Barkow Leda Cosmides and John Tooby Eds Leda Cosmides and John Tooby Mapping the mind Domain specificity in cognition and culture chapter Origins of domain specificity the evolu
65. t agents 2 working together to achieve a shared purpose 3 through a division of labor 4 integrated by decision processes 5 continuously through time An organization consists of patterns of behavior and interaction that are relatively sta ble and change slowly over time Sufficiently competent agents humans have the ability to organize themselves nat urally This is a direct result of the need to coordinate the work divided among the agents In other words organizing is a basic activity of competent agents If agent organizations emerge naturally why should we bother to design them The standard answer for human organizations is that properly designed organizations perform better than those that emerge naturally 10 It is reasonable to expect this answer to hold for software agent organizations as well 4 1 Choice of organization I will in the following try to explain the features of my choice of organization and why I have chosen it Given an organization design there must be a mechanism to 34 4 1 CHOICE OF ORGANIZATION 35 implement it both within each agent and in the agent organization architecture Three candidates and explanations to why two of them where rejected will be presented I did however implement two of them A comparison and test results may be found in Chapter 6 4 1 1 Single Agent Architecture The first model Figure 4 1 proposed is a single agent architecture and can hardly be called an
66. t distributions The repository used in the tests consists of 3256 cases and 79 packages The example target class I will use with the repository built from version 1 1 8 of the JDK consisting of method prototypes is illustrated below This is the same test case used in Case Based Support for Rapid Application Development 65 by Tessem et al class MyString tring char c tract boolean equals Object anObject t int length char data int length MyString Mys MyString int i abst abstrac abstrac t int charAt int i The example target class I will use with the repository built from version 1 3 of the JSDK consisting of method prototypes is class MyApplet String urlName URL myUrl int width int heigth MyApplet abstract URL getCodeBase abstract void play URL url abstract void resize int width int heigth The Agent settings are Number of Cases to be retrieved 8 Case match threshold 0 50 Average 48 6 1 RESULTS MANAGER WORKFORCE 49 6 1 Results Manager Workforce 6 1 1 JDK1 1 8 The eight classes retrieved for the MySt ring class from a repository built from ver sion 1 1 8 of the Java Development Kit were 1 java lang String 2 java text CompactCharArray 3 java util BitSet 4 java lang StringBuffer 5 java util Vector 6 java util GregorianCalendar 7 java util Hashtable 8 ja
67. the package also gets selected The target name similarity has to exceed 0 40 for package name similarity or 0 50 for class name similarity to select the package According to the Java Language Specification 26 p 106 it is recommended that naming conventions should be used in all Java programs so I also believe this is a reasonable solution In Katalagarianos and Vassiliou s work on software reuse 31 semantic similarity is the only feature taken into account in the matching If a Coordinator agent based on these calculations thinks that his Workforce may contain potentially reusable cases it passes a message to all the Workforce agents in the package 3 2 3 The Workforce Agents The individual case or Workforce agent possesses its own case description The de scriptions are created using Java s reflective facilities Java allows any class to be asked for its methods fields constructors inheritance information and other information at run time 63 Java s syntactic reuse construct is the import statement Java uses an environmental variable called CLASSPATH to establish where to search for classes 26 3 2 AGENT SUPPORTED CASE BASED RETRIEVAL 27 that are mentioned as import statements The agent supported case based retriever traverses the directories on the CLASSPATH environmental variable extracts all the feature information for each class in a pre processing step and stores that information ina case file associat
68. ther agents to believe or do something 2 3 Intelligence After decades the term intelligent has still not been defined or understood for artifi cial systems and applying it to agents may not be appropriate Most tend to regard the term agent and intelligent agent as equivalent Perhaps this is just an attempt to com municate that agents have more power than conventional approaches Agents could be e g in comparison to relational tables or objects thought of as somewhat smarter Or it could just be marketing hype However it would be fair to say that the notion of intelligence for agents could very well be different than for humans We are not cre ating agents to replace humans instead we are creating them to assist or supplement humans A different kind of intelligence then would be entirely appropriate If the science of artificial intelligence has made any contribution to human knowledge it is in confirming that intelligence is not some mystic vapor permeating men and angels but rather the effect of a set of general principles that can be understood and applied to the design of intelligent machines 36 p 75 Whatever the term intelligent agent means such agents will require a set of attributes and facilities E g an intelligent agents state must be formalized by knowledge i e beliefs goals desires intentions plans assumptions and be able to act on this knowledge It should be able to examine its beliefs an
69. tion of functional organization pages 85 116 Cambridge University Press Cambridge 1994 Lawrence A Hirschfeld and Susan A Gelman Eds Nachum Dershowitz Programming by analogy In R S Michalski J G Car bonell and T M Mitchell editors Machine Learning An artificial intelligence approach volume II chapter 15 pages 393 421 Morgan Kaufmann Inc Los Altos California 1986 Agent Working Group ed James Odell Agent technology green paper Work ing paper v0 91 http jamesodell com publications html 2000 Riza Cenk Erdur Oguz Dikenelli and Halil Sengonca A multiagent system for searching and retrieving reusable software components In 14th International 78 REFERENCES 79 16 17 18 19 20 21 22 23 24 Symposium on Computer and Information Sciences Kusadasi Izmir Turkey 1999 Luger G F Cognitive Science The Science of Intelligent Systems Academic Press San Diego and New York 1994 Carmen Fern ndez Chamizo Pedro Antonio Gonz les Calero Mercedes G mez Albarr n and Luis Hern ndez Y nez Supporting object reuse through case based reasoning In Ian Smith and Boi Faltings editors Advances in case based reasoning third European workshop EWCBR 96 Lecture Notes in Arti ficial Intelligence Springer Verlag 1996 Carmen Fern ndez Chamizo Pedro A Gonz lez Calero Luis Hern ndez Y nez and Alvaro Urech Baqu Case based retrieval of software c
70. va text CompactIntArray It took 1 min and 4 sec to run on the repository built from version 1 1 8 of the Java M Development Kit The repository consists of 1581 cases and 78 packages 6 1 2 JSDKI1 3 The eight classes retrieved for the MyApplet class from a repository built from ver sion 1 3 of the Java M Standard Development Kit were 1 javax swing JApplet 2 java applet Applet 3 javax swing JEditorPane 49 50 CHAPTER 6 COMPARISON AND TEST RESULTS 4 javax swing JTextPane 5 javax swing text html HTMLDocument 6 javax swing text html StyleSheet 7 javax swing JDialog 8 javax swing JTextComponent It took 3 min and 11 sec to run on the repository built from version 1 3 of the Java Standard Development Kit The repository consists of 3256 cases and 79 packages 6 2 Results Manager Coordinators Workforce 6 2 1 JDK1 1 8 The eight classes retrieved for the MySt ring class from a repository built from ver sion 1 1 8 of the Java Development Kit were exactly the same as under the Man ager Workforce Architecture 6 1 It took 49 sec to run on the repository built from version 1 1 8 of the Java M Devel opment Kit The repository consists of 1581 cases and 78 packages 6 2 2 JSDK1 3 The eight classes retrieved for the MyApplet class from a repository built from ver sion 1 3 of the Java M Standard Development Kit were 1 java applet Applet 2 javax swing JApplet 50 6 3 TEST R
71. vises them from the rest of the organization s workforce A coordinator leads each department and the Manager leads the whole organization Each member has the potentiality of surviving and succeeding but also the possibility of replacement and death They com municate by sending messages to each other and the command line is top down Chap ter 4 There are many other areas than autonomy and interaction that can differentiate the agent based approach from traditional OO 45 but I think the most important distinc tion is the computational model used to solve the problem In most cases the problem can be solved both ways The resulting program may look just the same externally but agent based programming should be able to deal with complexity and adaptabil ity better Agents can be viewed as an extension of objects An object is something that encapsulates its identity who its state what and its behavior how An active object encapsulates its own thread of control when An agent may have all of these but it must have something else A common approach is that an agent encapsulates why it does something Objects have a specific contract that defines their behavior 6 2 3 INTELLIGENCE 7 while agents like humans have only a generic interface to interact with others and the environment Contracts between agents are negotiated at a meta level They are not implicit in the interface Agents like humans can only attempt to get o
72. ween objects Standard AI knowledge representation such as frames semantic networks and if then rules may easily and naturally be implemented in Java 2 6 Bringing it all together Following AI method I have identified a main task for the agents and system as a whole This task is to retrieve software components for reuse The task can be viewed as demanding intelligence since the location of such code requires search recognition and comparison of code features and which if it should have been carried out by a human would demand knowledge experience and skill to accomplish Recalling Pattie Maes definition from Section 2 5 2 an agent is a process that lives in the world of computers and computer networks and can operate autonomously to fulfill one or more tasks 38 13 14 CHAPTER 2 THEORY AND METHOD This definition describes the features an agent has to possess but in relation to ASCRARAD it also expresses why agents are particularly suitable in this type of environment The concept of agents being autonomous makes it possible for the user to concentrate fully on her task Agent orientation is a great computational model for creating autonomous systems and I believe that agent support could become excellent help for the user Case based Reasoning CBR is a problem solving paradigm that uses knowledge of relevant past cases to interpret or to solve a new problem case The technique con stitutes a natural approach t
73. y 1997 34 Susan E Lander Distributed Search and Conflict Management Among Reusable Heterogeneous Agents PhD thesis University of Massachusetts 1994 35 C G Langton Artificial Life An Overview MIT Press Cambridge MA 1995 36 George F Luger and William A Stubblefield Artificial Intelligence Structures and Strategies for Complex Problem Solving Addison Wesley 1997 37 Pattie Maes http pattie www media mit edu people pattie homepage html 38 Pattie Maes Interacting with virtual pets digital alter egos and other soft ware agents Doors of Perception II Conference Amsterdam November 1994 http www doorsofperception com doors doors2 transcripts maes html 39 N A Maiden and A G Sutcliffe Exploiting reusable specifications through anal ogy Communications of the ACM 35 4 55 64 1992 40 M Minsky Society of Mind Simon and Schuster New York 1985 41 Kanth Miriyala and Mehdi T Harandi Automatic derivation of formal software specifications from informal descriptions JEEE Trans Software Engineering 17 10 1126 1142 1991 42 Stein Inge Morisbak Samarbeidende cbr agenter Assignment written in the course Artificial Intelligence iv281 at the Department Of Information Science University of Bergen May 1999 43 A Newell and H Simon Human Problem Solving Prentice Hall Engelwood Cliffs N J 1972 81 82 44 45 46 47 48 49 50 51
74. yland USA December 1995 Workshop on Intelligent Information Agents 59 Herbert A Simon Artificial intelligence an empirical science Artificial Intelli gence 77 1 95 177 1995 60 George Spanoudakis and Panos Constantopoulos Analogical reuse of require ments specifications A computational models Applied Artificial Intelligence 10 4 281 306 1996 61 Sun Microsystems Mountain View CA Java M 2 Platform Standard Edition v 1 2 2 API Specification http java sun com j2se 1 2 docs api 83 84 REFERENCES 62 Sun Microsystems Mountain View CA Java M Platform 1 1 API Specification http java sun com products jdk 1 1 docs api packages html 63 Sun Microsystems Mountain View CA Java 2 SDK Standard Edition Docu mentation 1999 http java sun com products jdk 1 2 docs 64 Bjgrnar Tessem and Solveig Bjgrnestad Analogy and complex software model ing Computers in Human Behavior 13 4 465 486 1997 65 Bjgrnar Tessem R Alan Whitehurst and Christopher L Powell Case based support for rapid application development In SEKE 99 Proceedings of the 11th International Conference on Software Engineering and Knowledge Engineering pages 192 196 Skokie Illinois 1999 Knowledge Systems Institute 66 R Alan Whitehurst Systematic Software Reuse Through Analogical Reasoning PhD thesis University of Illinois Urbana Champaign IL 1995 84

Download Pdf Manuals

Related Search

Related Contents

Bosch LTC0455/11 surveillance camera  Disk Array Controller Configuration Utilities Installation Guide and  Porting NetBSD/evbarm to the Arcom Viper  Installationsanweisung DA GL55Z/ 60Z/ 90Z  Logitech M515 User's Manual  SUPERDISK DRIVE UNIT SUPERDISK LAUFWERK LECTEUR  Targus SoundUP™  Pioneer AVIC-80DVD GPS Receiver User Manual  La maison de la culture de Firminy  

Copyright © All rights reserved. Failed to retrieve file