Individual project final presentations 2009, semester 1

Transcription

1 Individual project final presentations 2009, semester 1

2 Interactive Selection Nor Nazrina Mohamad Nazry, Tom Gedeon COMP8780 Information & Human Centered Computing Project Master of Computing July

3 Selection of Preferred Pictures Examples of Pictures ( Mondrians ) 2

4 Using Different Approaches Mouse Interface Driving Interface 3

5 Modification of Mouse Interface Mouse Interface Previous Work Mouse Interface Current Work 4

6 Modification of Driving Interface Driving Interface Previous Work Driving Interface Current Work 5

7 Experiment Designs Experiment (Description) Mondrian(s) Displayed at One Time Mouse Interface Driving Interface Users Current (3:3) (70 sets x 3 Mondrians) (70 sets x 3 Mondrians) Current (3:1) (70 sets x 3 Mondrians) (210 sets x1 Mondrian) Current (20:3) (10 sets x 20 Mondrians) + (1 set x 10 Mondrians) (70 sets x 3 Mondrians) Previous (20:1) (20 Mondrians originally and less one for each new set) (210 sets x1 Mondrian) Total of Mondrians Displayed for Each Interface: 210 Mondrians 6

8 Experiment Layout 7

9 Results Analysis Types of Analysis Qualitative analysis (speed, tediousness, ease of selection, ease of learning) Quantitative analysis (no. of Mondrians selected, learning speed, average speed) Analyzed under 4 Factors Interface the user rates (mouse, driving) The group the user is in ( Driving then Mouse or Mouse then Driving ) User s computer skill (good or average or bad) User s driving skill (good or average or bad) 8

10 Results Rank of User Experiment (Description) Rank of User Computer Skill Driving Skill Good Average Bad Good Average Bad User Current (3:3) Current (3:1) Current (20:3) Previous (20:1)

11 Results Qualitative Analysis User Rated Tediousness by Interface Current (3:3) User Rated Lack of Tediousness Mouse Driving Lack of Tediousness 10

12 Results Quantitative Analysis Average Speed by Interface Current (3:3) Speed per Mondrian Mouse Driving Average Speed 11

13 Conclusion Results Discussion Same rating of qualitative analysis driving interface is as fast and intuitive as mouse interface (Previous work driving interface was slow) Mouse interface a bit more tedious (Previous work driving interface was much more tedious) Future Work Comprehensive test Redesign the layout of mouse interface Improve the quality of driving interface 12

14 Intermission

15 Final Project Presentation VARIATION OF DEFECTS AND COMPLEXITY IN MULTIPLE VERSIONS OF OPEN-SOURCE PROJECTS Kshitij Shukla (U ) Supervisor: Clive Boughton June, 2009

16 BACKGROUND Card and Glass Structural oriented software Pre delivery defects Normi and Clive Object oriented software Post delivery defects

17 DEFINITIONS Structural Complexity (Fanout or CBO) Fanout is the number of local flows out of a module plus the number of data structures that are used as output. Data Complexity (Numbers of parameters) Data complexity of a module is the number of data items (variables) it is expected to process. Procedural/ Cyclomatic Complexity It s the measure of control flow within the module. System Complexity Structural Complexity + Data complexity

18 METRICS TOOL VALIDATION Tool tested: Chidamber & Kermerer Java Metrics (CKJM) JStyle JHawk Resource Standard Metrics (RSM) Outcome/Result JHawk gives Fanout, CBO and number of parameter values closer to our manual calculation. JStyle gives value closer to our value for Cyclomatic complexity

19 SOFTWARE SYSTEM DATA ANALYSIS Step 1: Selecting and downloading versions of 5 open source Java software Feeding source code of different software version to JHawk/JStyle Collecting post-delivery defects information Step 2: Data analysis to find trend and relationship for all data collected

20 System Complexity SOFTWARE SYSTEM DATA ANALYSIS (1) System Complexity Version Number HTMLUnit LaTeXDraw Saxon DataCrow ZK

21 Number of classes SOFTWARE SYSTEM DATA ANALYSIS (2) 1600 Number of Classes Version Number HTMLUnit LaTeXDraw Saxon DataCrow ZK

22 System Complexity per class SOFTWARE SYSTEM DATA ANALYSIS (3) 30 System Complexity per class Version Number HTMLUnit LaTeXDraw Saxon DataCrow ZK

23 SOFTWARE SYSTEM DATA ANALYSIS (4) Software Product Reported post delivery Defects v/s Versions Reported post delivery v/s System Complexity Number of Downloads v/s Versions Reported post delivery Defects v/s Number of Downloads Reported post delivery Defects per Number of Downloads v/s Versions ZK Positive Positive Positive Positive Positive DataCrow Positive Negative Positive Negative Negative Saxon Positive Positive Positive Negative Positive LaTeXDraw Negative Negative Positive Negative Negative HTMLUnit Positive Positive Positive Positive Negative Majority

24 Cyclomatic Complexity per class SOFTWARE SYSTEM DATA ANALYSIS (5) 30 Cyclomatic Complexity per class Version Number HTMLUnit Saxon DataCrow ZK

25 Cyclomatic /System Complexity SOFTWARE SYSTEM DATA ANALYSIS (6) 2.5 Cyclomatic Complexity /System Complexity Version Number HTMLUnit Saxon DataCrow ZK

26 CONCLUSION & FUTURE WORK JHawk and JStyle gives value closer to our manual calculation. Successfully shown the use of metrics-calculating tools to find correlation between System complexity and different version releases Reported number of defects and different version releases System complexity and reported number of defects Cyclomatic complexity and different version releases Future work includes extension of the work, by considering more than five open source Java software.

27 THANK YOU QUESTIONS???

28 Intermission

29 Phone Bill Analyser COMP8780 Information and Human Centred Computing Project Implementation Project Stephen Bowman - u Project supervisor: Peter Christen

30 Context Diagram Customer Digital Phone Bill Report - Plan Rankings and Costs Phone Bill Analyser Plan Details Telco Websites

31 Problems Finding the Best Plan Too many options 302 Australian telephone service providers [1] Assuming 5 plans per service provider, 1,510 plans Too many calculations Assuming 50 calls per month, ~75,500 calculations Problem is even worse when plan combinations are considered Limitations of existing systems Unclear/ tricky conditions [1] Telecommunications Industry Ombudsman

32 Development Methodology Higher Requirements Level of Abstraction Project Description Vision and Scope Use Cases Prototype Architecture Context Diagram Data Flow Diagrams Technology Map Specification for Specification for Integration Test System Test Design Class Diagram State Charts Candidate Selection Spec. for Unit Test Lower Project Start Software Artefacts Procedures Progress of Project Project Complete & System Developed

33 Example Report Page 1

34 Example Report Page 2

35 Still have a problem Still too many plans and plan combinations to calculate the customer s call costs on each plan The solution = a candidate selection model: Select candidate plans for a customer based on a summary of their call activity Consists of a set of rules of the form: If <conditional involving call activity> then add plan X to the candidate set The Phone Bill Analyser calculates actual costs for these candidates

36 Creating the Model

37 1. Create Training Data Call Activity Summary Plan in Candidate Set? % calls on weekend # calls P1 P Y Y Y N Y N N N

38 2. Generate Rules Use Weka [2] For each plan: Import training data Generate rules using Part rule generator Example rules (from above training data): If (% calls on weekend)<=10% then P1 = Y If (% calls on weekend)<=10% & (# calls <= 100) then P2 = Y [2] Witten & Frank 2005

39 3. Score Rules for Each Plan Actual Class Y N Predicted Class Y N

40 3. Score Rules for Each Plan Actual Class Y N Example: P1 confusion matrix Predicted Class Y N P1 Total Cost =

41 4. Convert Rules to Java For each rule with target Y, add the plan to the candidate set if the conditional is true If there are insufficient candidates after all rules have been tested, then add plans in order of decreasing total cost, until there are enough candidates

42 5. Use Model

43 Future Work Programmatic Weka interface Automate the extraction, correction and upload of plan data Extend testing and use Increase range of data mining techniques Research learning classification rules for multiple target attributes

44 Full References [1] Telecommunications Industry Ombudsman, 2009, Search Members List, members.htm, viewed 16 March [2] Witten, I & Frank, E 2005, Data Mining: Practical Machine Learning Tools and Techniques (Second Edition), Morgan Kaufmann, San Francisco, California.

45 Questions?

46 Intermission

47 OSGi in Distributed Robotic Systems Ben Coughlan 2009

48 u biq ui tous adj. existing or being everywhere, esp. at the same time; omnipresent: ubiquitous fog; ubiquitous little ants. dictionary.com het er og e nous adj. Consisting of dissimilar elements or parts; not homogeneous. dictionary.com Ben Coughlan June 2009

49 Ben Coughlan June 2009

50 Ben Coughlan June 2009

51 Ben Coughlan June 2009

52 Ben Coughlan June 2009

53 Ben Coughlan June 2009

54 Ben Coughlan June 2009

55 Ben Coughlan June 2009

56 Intermission

57 A Platform Independent GUI For The PeANUt Software Simulator Project Supervisor- Dr Peter Strazdins Project Developer- Avinash G Prasad 1

58 PeANUt PeANUt was created by the department of Computer Science for teaching purposes. PeANUt is a 16 bit simple software microprocessor with a 16bit arithmetic and logic unit, an addressing unit, an execution unit and its own primitive operating system. 2

59 Original PeANUt Screenshot 3

60 Key Issues Platform Compatibility Maintainability of the software Uses Tcl/Tk scripting language 4

61 Old PeANUt vs. New PeANUt Old PeANUt Functionality in C New PeANUt Functionality in C GUI in Tix Import Legacy code using JNI Tcl/Tk Scripting Old PeANUt GUI in JAVA New PeANUt 5

62 Java Native Interface Advantages Use the existing library that was previously written in another languages. Allows JAVA to access some hardware features Increasing the speed of execution 6

63 Java Native Interface Disadvantages Write Once Run Anywhere is not possible Run time errors debugging is difficult in native code An applet can not call a native method Security risk is potential 7

64 Class Diagram 8

65 Sample JNI Code //Java Code Example- Declare a native function public native int Cfunction(); // Java-Call the native method int n= Cfunction(); //C Declaration JNIEXPORT jint JNICALL Java_Example_ Cfunction(JNIEnv *jenv, jobject job) 9

66 Automated Testing Using Abott and Costello Running Scripts with Costello 10

67 New PeANUt 11

68 Thank You 12

69 Intermission

70 Comp8790 Project, 2009 Australian National Univeristy Name: Lin Chen ID: u

71

72

73

74 SNA module enabled

75

76

77

78 Example: Occupation options

79 Add This user (Test 6) to Daric Invitation would be sent for test6 to accept.

80

81

82

83

84 Click on button to download file

85

86

87

88 In sna.module: sna_cron() is called to rebuild sna data: function sna_cron() { include 'explore.php'; }

89 They are GML format Output when cron runs.

90

91

92 Intermission

93 Comp8790 Optimizing the VOSON e-research tool Chen Xiantong U Tutor: Dr Robert Ackland, Dr Uwe Zimmer

94 Comp8790: Introduction Virtual Observatory for the Study of Online Networks (VOSON) e-research tool is a web-based software incorporating web mining, data visualization, and traditional empirical social science methods.

95 Comp8790 Introduction One of the most essential processes of the VOSON system is data preparation and analysis. In this paper, we focused on two bottleneck: Data structure of mapping nodes. SQL querying of updating records. Evaluation Implementation Tool Evaluation Runtime Evaluation hash_map Data Stucture unordered_map Connection Pool Parallel Querying External Connection Benchmark Data integrity and consistency test Tool Benchmark Runtime Benchmark

96 Comp8790 Project Evironment Fedora Core 10 *.params.xml Original DB boost () createanalysisdb () createnodesvec_fromtable1 mysql 5.0 () createnodesvec_pagegroups gprof massif () savenodesvec_totable1 Analysis DB () createshadowdb Shadow DB

97 Comp8790: Evaluation System of evaluation: 2 cores system with Inter Pentium GHz, 512 MB memory. Database of evaluation: nano_nano_nano2 Used by Dr Robert Ackland to study the online presence of environmental activities organizations Map nodes Update records Tools of evaluation: gprof: profiling tool of running time and call graph. massif: profiling tool of memory consumption on stack.

98 Comp8790: Evaluation Runtime evaluation: main(): leaving createnodesvec_fromtable1(). Took -4- seconds. leaving createnodesvec_pagegroups(). Took-0- seconds. leaving savenodesvec_totable1(). Took -7- seconds % 10 0% leaving createanalysisdb_main(). Took -15- seconds. ULonline finished % 10 0% savenodesvec_totable1(): Updating records, uses 7 seconds. createnodesvec_fromtable1(): time(secs) % 10 0% 26.7% 10 0% Mapping records, uses 4 seconds. 0 m ain...fr om Table1...pagegroups...toTable1 Other s functions

99 Comp8790: Evaluation () createnodesvec_fromtable1 Evaluation of gprof: % time Total calls Name ()() std::less<int>::operator () std::_rb_tree<>::_m_lower_bound ()() std::_select1st<>::operator () std::_rb_tree<>::_s_key massif: Total memory usage: MB Peak statistics: 82.89% (17,833,318B) (heap allocation functions) malloc/new/new[] () gnu_cxx::new_allocator<node_info_>::allocate() (10,092,544B) 0x806D62F: ->46.91%

100 Comp8790: Implementation Implementation of nodes mapping: Disadvantages of map: Search, insert and delete in O(log n) time. Additional memory for leaves and color attribute. Advantages of hash table based containers. Search time independent of the number of elements. Insertions and deletions at constant average cost of O(1).

101 Comp8790: Implementation Implementation of nodes mapping: hash_map: #include <hash_map> using namespace gnu_cxx; typedef hash_map<int, int> NodesContainer; unordered_map: #include <tr1/unordered_map> typedef tr1::unordered_map<int, int> NodesContainer; -std = c++0x

102 Comp8790: Implementation Implementation of parallel querying: 2 modes of parallel programming: Shared-memory mode represented by OpenMP. Message-passing mode represented by MPI. Advantages of OpenMP: Optimized for loop operations. portable and scalable for both sequential and parallel programming.

103 Comp8790: Implementation Implementation of parallel querying: Connection Pool Connection Pool main External connections: extern mysqlpp::connection con; mysqlpp::query query = con.query(); conn0 conn1 () create if (omp_get_thread_num() == 0) { query.reset(); query << querybuff.str(); update_dbase(query); } conn2 () destroy ( firstprivate(i #pragma omp parallel for if_def OMP_H -fopenmp

104 Comp8790: Best solution Comparison of nodes mapping solutions: Name Time Memory map 15.5s 20.52MB hash_map 14.3s 20.44MB unordered_map 14.5s 20.16MB Comparison of parallel querying solutions: Name Time Memory single 15.5s 20.52MB connection pool 17.2s 21.10MB external connection 14.5s 20.92MB

105 Comp8790: Benchmark Data integrity and consistency MySQL tools checking: - run the original voson code to create the old DB - mysqldump -u test -px --skip-extended-insert nano_test_nano2_ul > old.sql - remove the timestamp using perl - do the same for the new voson code to create new.sql - then compare old.sql and new.sql using diff old.sql new.sql

106 Comp8790: Benchmark Benchmark by tools gprof: % time Total calls Name ()_ node_info_::node_info ()<> gnu_cxx:: normal_iterator () savenodesvec_totable ()<> std:: copy_move_a2 int* massif: Total memory usage: MB Peak statistics: 80.89% (17,334,602B) (heap allocation functions) malloc/new/new[]

107 Comp8790: Benchmark 18 time(secs) main..._fromtable1...totable Original hash_m ap NO OMP hash_m ap OMP solutions

108 Comp8790: Discussion Further discussion The reasons of poor performance gained by connection pool. The reasons of not using parallel programming in nodes mapping. The reasons of unexceptional performance gain. Improvement of future project

109 Intermission