Home

Multi-Dimensional Explorer - Das IICM

1. L L Press to toggle layo Gender Country Poltics Relationship Network GMT Walposts Groups Photos Total friends Press cto cle b Paired Parallel Coordinates Figure 3 23 Situvis Clear et al 2009 is an open source software package developed within the Processing programming environment featuring parallel coordinates visualisation Although envisioned as a tool for modelling user behaviour patterns it can be used for all kinds of appli cations a shows what the software currently looks like Previous versions of Situvis as shown in b pair parallel coordinates with a graph visualisation based on the observation that parallel coordinates are not particularly good at visualising relations between individual records The graph visualisation does exactly that its nodes represent records its edges represent relations between them Selections in both visualisations are synchronised with turquoise indicating a selection Turquoise nodes in the graph correspond to turquoise lines in the parallel coordinates display Purple nodes and lines represent records directly related to the selected ones This im age shows an example of a social network data set in which each record represents a user on Facebook Note that both images shown here have been inverted to increase legibility 41 42 3 Parallel Coordinates MM Picviz Frontend BETA AND SLOW File Edit Selection View Help
2. 299 99063 12 0 17 280449 4 0 17 296621 0 0 0 0C length breadth depth draft speed blending_coefficient object status E hist Y tooltips line W Brush Fuzziness 20 rendering quality Figure 3 21 parvis is an open source Java desktop application for general purpose parallel coor dinates visualisation displaying here the well known cars data set Ramos and Donoh 1983 This figure illustrates progressive rendering one of parvis special features to enhance speed and usability While an axis is being moved by dragging it horizontally with the mouse the visualisation is updated immediately without anti aliasing It is then progressively replaced by anti aliased lines while the application stays responsive allowing the user to continue work with out interruption The bar at the bottom left of the window shows the current rendering progress 38 3 Parallel Coordinates ParallAX D Uni masterarbeit software parallax package parallax dat CURRENCY DAT Fie Anows Query Vas Types vlew Scales Window aNalysis Classifiers sEtup Help Figure 3 22 Parallax a commercial closed source desktop application featuring parallel coor dinates This figure shows a combination of two different query types on a financial data set Records appearing in green have been selected by an angle query on the Year axis blue ones have been selected by an interval query on the Yen axis
3. Q Q gs Disperse Axis Hide Show Only Show All 7 Anti Aliasing Time source Destination I Sport port I setts Jopkts 2 soytes y fobytes change aja 1 2 Layers default af Time source Destination Sport Dport Spkts Dpkts Sbytes Dbytes 22 30 22 192 168 1 160 192 168 1 158 50541 5900 22 30 22 192 168 1 160 192 168 1 158 50541 256 22 30 22 192 168 1 160 192 168 1 158 50541 3306 22 30 22 192 168 1 160 192 168 1 158 50541 443 22 30 22 192 168 1 160 192 168 1 158 50541 445 a 3 Q g 22 20 22 197 16R 1160 192168 1158 snsa1 a Figure 3 24 The standard GUI frontend of Picviz an open source C library aimed specifically at applying parallel coordinates visualisation to network security The data set shown here is scanned network traffic Grey lines represent selected records In addition dashed rectangles are drawn around selected records dimension wise more than once an axis cannot however be hidden completely other than by replacing it with a different axis Instead of selecting records by defining intervals on or between axes Picviz employs a generalised approach one can draw a selection rectangle with the mouse to enclose desired polylines Selected records can then be coloured individually Other than that the colours used in the display cannot be changed Sets of records can however be saved as layers and subsequently be hidden from the view Thi
4. Cairo 2010 Cairo http cairographics org Cited on page 40 Cawthon Nick and Andrew Vande Moere 2007 The Effect of Aesthetic on the Usability of Data Visualization In Proc 11 IEEE International Conference on Information Visualisation IV 2007 pages 637 648 ISSN 1550 6037 doi 10 1109 IV 2007 147 http web arch usyd edu au andrew publications iv07b pdf Cited on page 6 Bibliography 121 Celik Tantek Bert Bos lan Hickson and H kon Wium Lie 2010 Cascading Style Sheets Level 2 Revision I CSS 2 1 Specification Candidate recommendation World Wide Web Consortium http www w3 org TR CSS21 Cited on page 64 Chambers John M William S Cleveland Paul A Tukey and Beat Kleiner 1983 Graphical Methods for Data Analysis Duxbury Press ISBN 053498052X Cited on page 8 Clear Adrian K Ross Shannon Thomas Holland Aaron J Quigley Simon A Dobson and Paddy Nixon 2009 Situvis A Visual Tool for Modeling a User s Behaviour Patterns in a Pervasive En vironment In Proc 7 International Conference on Pervasive Computing Pervasive 2009 pages 327 341 doi 10 1007 978 3 642 01516 8 22 http situvis com Cited on pages 40 41 and 80 Cleveland William S and Robert McGill 1984 The Many Faces of a Scatterplot Journal of the American Statistical Association 79 388 pages 807 822 ISSN 0162 1459 http www jstor org stable 2288711 Cited on page 7 Cook Dianne and Deborah F Sway
5. Category Created Importance Quality Used but unselected record Unused record Dragging from 0 6524999737739563 0 27250000834465027 to 0 7574999928474426 0 45749998092651367 b Some records are unused or selected Figure 3 19 The parallel coordinates component in MDVS Hackl 2010 visualising a test data set with dimensions Category Created Importance and Quality a shows all records are used and unselected However as b shows records can also be in different states When unused because of a restricted dimension axis they appear grey When in use they can be either selected blue or unselected black 3 5 Software Applications ings around the Globe 2009 mis Geographic Locations c Thematic Similarities Continents World North America South America Europe Africa Asia Oceania Graphical view Figure 3 20 Macrofocus InfoScope a commercial visualisation application written in Java aimed at data with a geographical dimension Depending on the nature of the data set the parallel coordinates display is synchronised with a geographical map In this example the data is about cities spread across the globe the lines highlighted in the parallel coordinates display correspond to the cities shown on the world map The city of Milan corresponds to the red dot in the centre of Europe and to the red selected polyline in the parallel coordinates display Likewise Vienn
6. a Parallel Coordinates with no Clustering WM Eile View Brushes Tools Window Help ES A zoomx 100 ZoomY 100 ZoomDrecton x IR E CR ca pre s i ca gud p we 2 1765 6 47 5 785 341 5 14 92 2547 17 07 352 1105 237 174 OLEOLE Lu Pao b Parallel Coordinates with Clustering Figure 3 6 Hierarchical parallel coordinates display in XmdvTool 2010 building on the idea of Fua et al 1999 a is a normal parallel coordinates visualisation of the cereals data set provided by Velleman 1996 b shows the results of applying a clustering method to the data set 3 3 Variations 19 the display Whenever one line is added opacity of neighbouring lines is increased while opacity of all lines is reduced Allowing users to pause the resulting animation allows them to select their desired level of detail Animation is noted as an important visualisation technique also by Ellis and Dix 2007 who give an overview of different approaches to mitigate the problem of cluttered displays in information visualisation and conclude that better results are achieved by a combination of different methods such as animating the clustering process Johansson et al 2005b use animation as an informa tion channel of its own using motion in a clustered parallel coordinates visualisation to express additional information about a cluster of records A high number of dimensions rather than just recor
7. 5 6 namespace mde 8 namespace presentation a 10 class TestVisualisation public Visualisation 11 12 public 13 TestVisualisation DataViewConstructionParameters const amp parameters 14 15 private 16 virtual void partiallyCreateVisualisation int record_index 17 Es 18 19 20 endif Listing B 1 The header file for Testvisualisation The class is derived from Visualisation and implements its pure virtual function partiallyCreateVisualisation The DataViewConstructionParameters parameter is just passed to the base constructor TestVisualisation itself does not use it AHA U A U N oo 10 11 12 13 14 15 16 B 3 Implementing Visualisations 113 include test_visualisation h namespace mde namespace presentation TestVisualisation TestVisualisation DataViewConstructionParameters const amp parameters Visualisation parameters void TestVisualisation partiallyCreateVisualisation int record_index Listing B 2 The source file for TestVisualisation The constructor passes parameters to the base constructor while partiallyCreateVisualisation currently does nothing This is skeleton code for future visualisation derived MDE classes Now that the class exists its existence must be made known to other components of MDE DataView FactoryGroup contains a list of all Dataview derived concrete classes Including Test Visualisation there will automatically make it appear in MDE
8. 50 I Rotate Labels 57 _F Meanline I Polyines Hstograms I Left I Right Histogram Bins 5 Options 73 records 73 used 1 selected 52 dimensions Figure 3 14 One of the records in a parallel coordinates visualisation is selected In this exam ple building on the example in Figure 3 13 one record stood out between the two apparently correlating dimensions Normal Local Rent Medium and Time Required for 1 kg of bread minutes due to the direction of its line It has been selected here highlighted in blue and revealed to be that of Caracas Therefore Caracas is an example of a place where rents are much higher compared to other cities in relation to the time required to earn enough money for 1 kg of bread The selected record is shown in blue so that the user can distinguish between selected and unselected records 3 4 Using Parallel Coordinates 29 Figure 3 15 Angular brushing is used to select records in parallel coordinates by defining an angle interval between two axes On the right there is an illustration of interval definition the upper red line has an angle of 45 degrees from the axis lines the bottom red line an angle of 90 degrees Applied between axes A and B shown on the left side the angle between every polyline segment and the axis lines is considered the red segments match the filter because for them the angle is between 45 and 90 degrees so they are considered se
9. J I Meantine I Polyines Histograms I Left I Right Histogram ins 5 H Options 73 records 73 used 0 selected 52 dmensions Figure 3 10 Initially a parallel coordinates visualisation usually shows all axes The cities data Set used here UBS 2009 contains 52 dimensions The user will want to remove some if not most of them from the view cities2009 csv Multi Dimensional Explorer Parallel Coordinates SvGExport zoom 100 Polyine Opacty 50 f M Rotatetabels 30 f I Meantine Y Polyines Histograms I Left I Right Histogram Bins 5 Options I 73 records 73 used 0 selected 52 dimensions Figure 3 11 In contrast to Figure 3 10 all but 5 of the 52 dimensions in the Cities Data Set have been hidden by the user leaving City Names Paid Vacation Days Per Year Time Required for 1 kg of Bread Time Required for 1 kg of Rice and Normal Local Rent Medium 26 3 Parallel Coordinates 3 4 2 Moving Axes Once the set of dimensions has been reduced to a suitable subset ordering becomes an issue Axes in par allel coordinates are restricted to two neighbours at least in their traditional 2D form see Section 3 3 2 for a variation of parallel coordinates overcoming this limitation and correlations between dimensions can be explored only if the corresponding axes are placed next to each other This calls for a means to move axes to obtain a desired ordering Fi
10. Table 4 3 The classes enums typedefs and free standing functions of MDE s Data module 4 2 Modules 63 FileReader readMatrix path Matrix amp bool doReadMatrix path Matrix amp bool A CsvFileReaderWithDialog CsvFileReader gt doReadMatrix path Matrix amp bool doReadMatrix path Matrix amp bool CsvFileReaderDialog wxDialog gt CsvFileReaderDialog path getMatrix const Matrix Figure 4 4 This class diagram shows how the part of MDE responsible for importing CSV files is implemented data input FileReader is an abstract base class to read any kind of files into data Matrix instances data input CsvFileReader is derived from it implementing its pure virtual doReadMatrix member function application CsvFileReaderWithDialog adapts data input CsvFileReader for GUI specific use implemented in terms of application CsvFileReaderDialog the Import Wizard from the user s point of view This designs completely separates parsing and interpreting CSV files a concern of module Data Input from the provision of related GUI elements to the user the concern of module Application 4 2 3 Data Input This module is a sub module of Data It contains tools to read data from external sources and use it in the creation of new mde data Matrix instances The templatised non default constructo
11. StarlD Hip HD HR Gliese BayerFlamsteed ProperName RA Dec Distance Mag AbsMag Spectrum Colorln SVG Export Zoom 100 Polyine Opacity 50 T Rotate Labels 30 f I MeanLine I Polyines Histoorams I Left I Richt Histogram Bins 5 Options Record 2078 5985 8582 236890 1 8458751 55 919052 49 285362 8 96 5 4964102 GO 0 78 3000 records 3000 used O selected 14 dimensions b After focussing Figure 3 18 Focussing is a more user friendly way than zooming to discern individual records in a crowded parallel coordinates visualisation a shows 3000 records and 14 dimensions from the stellar data set by Nash 2006 On the top area between the second and third axis one of the polyline segments stands out but it is impossible to follow its path in the rest of the display because it is overlapped by thousands of other polylines On b the mouse pointer is hovering above the line segment between the second and third axis which temporarily focusses the polyline highlighting it in red and showing its individual data values 3 5 Software Applications 33 Name Characteristics MDVS Intuitive usage InfoScope Support for geographical dimensions parvis Performance gains by two phase rendering Parallax Visual classification selection mechanisms Situvis Conflict analysis pairing with graphs Picviz Aimed at network security GGobi Extensive statistical tools XmdvTool Time
12. a geographical map of the data Figure 3 29 shows the OECD eXplorer visualising statistics about un employment rates with parallel coordinates and a linked geographical map of Central European regions Standard parallel coordinates features include changing polyline opacity flipping axes free reorder ing hiding axes and setting interval filters on any axis Colours can be freely customised The minimum and maximum value can be specified for every axis As for special features there are histograms on the axes their width depending on the number of records in the corresponding axis interval Groups of records are selected by clicking on histograms There is also a mean line that is a special polyline representing the average value of every dimension On the negative side axes cannot be repeated and there are no tooltips for polylines regardless of whether they are selected or not 3 5 12 Protovis Protovis Bostock and Heer 2009 is a toolkit implemented entirely in JavaScript and SVG Mozilla Developer Network 2011 Ramani et al 2008 Its creators maintain that its high level of abstraction compared to many other low level graphics libraries is a distinct advantage for the development of infor mation visualisation software Protovis also includes parallel coordinates as shown in Figure 3 30 Given that Protovis is meant to run in a web browser and makes extensive use of scripting perfor mance is an issue User interaction is ve
13. s main window tool bar and menu bar In order to do so a factory class producing Test Visualisation instances derived from DataViewFactory is required The factory class will be called Test VisualisationFactory Again header files and source files as described above must be added to the Presentation project of the MDE solution The resulting factory code is shown in Listings B 3 and B 4 withInstanceIndex is used so that MDE can keep track of the number of instances of a particular product The list of option names created in the constructor and passed to the base factory class indicates which options TestVisualisation instances are going to use It is important to include rendering_backend in the list because every visualisation needs that option At this point the factory class can be added to DataViewFactoryGroup by including its header file in data_view_factory_group cpp and inserting a line in its constructor pushing back a dynamically allo cated instance of Test VisualisationFactory to factories The factory instance is never destroyed This is on purpose Developers should not falsely assume a memory leak here and attempt to needlessly delete the factories because this can lead to dangerous order of destruction issues Listing B 5 shows what the constructor of DataViewFactoryGroup looks like after the modification MDE now recognises the existence of the new visualisation in its user interface and displays appro priate butto
14. vin my ing Ur Ye oy ey 7 7 inay Meg rng log ey ry Yan 7 SVG Export zoom 100 f Polyine Opadty s0 J IV RotateLabels 30 fJ F Meantine Y Polyines Histograms I Left IV Right HstogramBins 5 _ Options Records ee imi xi Showallrecords Show only used C Show only used and selected Options se No Cities cross purchasing power Net purchasing power Prices exc 15 Chicago 97 9 96 3 82 0 18 Dubai 45 1 62 3 84 9 19 Dublin 90 7 106 4 92 7 20 Frankfurt 929 34 6 90 8 Bank credit clerk USD per year r Engineer USD per year heen bs ms iaa SVGExport zoom 100 x Axis Engineer USD per year lt Axis Bank credit derk USD per year y 73 records 22used 10 selected 52 dimensions Figure 4 10 MDE visualising the Cities data set UBS 2009 in three different ways The upper subwindow shows a parallel coordinates view of the data with some dimensions hidden There are filters defined on some axes turning a number of records unused and shown in grey Blue records are selected Record states are synchronised across different visualisation windows as is evident in both the scatter plot and the tabular records subwindow which use the same colours for the states The parallel coordinates subwindow features axis histograms and a mean line The first axis is highlighted in red meaning that the mouse pointer is currently hovering over it
15. 22 23 24 25 26 27 28 29 30 31 B 3 Implementing Visualisations 115 include test_visualisation_factory h include test_ visualisation h include data_view_construction_parameters h include lt boost assign list_of hpp gt namespace mde namespace presentation TestVisualisationFactory TestVisualisationFactory DataViewFactory boost assign list_of records_colour selected_records_colour unused_records_colour rendering_backend background_colour DataView TestVisualisationFactory doCreateDataView DataViewConstructionParameters const parameters return new TestVisualisation withInstanceIndex parameters productCounter std string TestVisualisationFactory getName const return Test Listing B 4 The source file for TestvisualisationFactory The list of option names in the constructor of the factory indicates which options Test Visualisation instances are going to use DataViewFactoryGroup DataViewFactoryGroup factories_ factories_ push_back new ParallelCoordinatesVisualisationFactory factories_ push_back new ScatterPlotVisualisationFactory factories_ push_back mew RecordTableDataViewFactory factories_ push_back new DimensionTableDataViewFactory factories_ push_back new TestVisualisationFactory Listing B 5 The existence Of TestVisualisationFactory is m
16. A Tutorial and Reference Addison Wesley Pro fessional ISBN 0201379260 http www josuttis com libbook index html Cited on page 54 Keim Daniel A and Hans Peter Kriegel 1994 VisDB Database Exploration using Multidimensional Visualization IEEE Computer Graphics 4 Application Journal 14 5 pages 40 49 doi 10 1109 38 310723 Cited on page 45 Keim Daniel A Hans Peter Kriegel Ankerst and Juraj Porada 2002 VisDB A Visual Data Mining and Database Exploration System http infovis uni konstanz de research projects VisDB visdb html Cited on pages xi 45 and 80 Khronos Group 2011a OpenGL The Industry Standard for High Performance Graphics http www opengl org Cited on pages 6 44 46 53 55 68 and 83 Khronos Group 2011b Survey Of OpenGL Font Technology http www opengl org resources features fontsurvey Cited on page 86 Kohonen Teuvo 1997 Self Organizing Maps Second Edition Springer ISBN 3540620176 Cited on page 17 l Agefi 2003 Classement des villes http www stadtzug ch dl php de 20031007081610 Classement_Villes2003 paf Cited on page 36 Ledermann Florian 2003 parvis parallel coordinates visualisation http www mediavirus org parvis Cited on pages 36 and 80 Lex Alexander 2008 Exploration of Gene Expression Data in a Visually Linked Environment Master s Thesis Graz University of Technology Austria http www icg tugraz at Members alex lex_
17. FTGLPixmapFont is used in the overload taking no such parameter for the drawing of non rotated text because its implementation is faster In both cases an instance of the corresponding FTGL class is dynamically allocated via a boost shared ptr This is a costly operation which should not be repeated every time text is drawn Therefore performance is optimised by caching the instances inserting them into a std map which maps font filenames to either boost shared_ptr lt FTGLTextureFont gt Or boost shared_ptr lt FTGLPixmapFont gt Having now at its disposal an instance of the proper FTGL font class doDrawText just uses its Face Size and Render high level methods to draw text on the current OpenGL visualisation For FTGLTextureFont the class used for rotated text the implementation is slightly more complex because OpenGL matrix transformations are involved First the identity matrix is loaded by calling the OpenGL function glLoadIdentity It is then modified by glortho such that the desired text coordinates become the point of origin Afterwards the projection matrix is rotated using glRotatef At this point the rotated piece of text is drawn When the operation has finished it restores the previous projection matrix Figure 5 3 shows the geometry of text rotation 5 2 Rendering TrueType Fonts in OpenGL Ney 7 Figure 5 3 MDE uses OpenGL s transformation matrix mechanism to rotate text drawn with t
18. Holder to allow it to access the data set and options Synchroniser to grant it access to program options How visualisation takes place and what the user can do in a subwindow is the concern of module Presentation Table 4 2 gives an overview of all classes functions and enumerations used by module Application 58 4 Multi Dimensional Explorer Name Purpose Application Main application class for wx Widgets CsvCharacterControl Character option control for CsvFileReaderDialogue CsvFileReaderDialogue Dialogue window for CSV file import CsvFileReaderWithDialogue Creates a data Matrix from a CSV file with user interaction DataViewMdiChildFrame wxMDIChildFrame used as parent window of presentation DataView ainMenuBar Menu bar for Mainwindow ainStatusBar Status bar for MainWindow ainToolBar Tool bar for Mainwindow ainWindow Top level window of MDE ainWindowItemInformation Appearance information for Mainwindow menus and toolbar ainWindowItems Group of MainWindowItemInformat ion instances OptionColourControl wxColourPickerCtrl replacement OptionColourTextControl Specialised text control for colour codes OptionControl Control on an OptionsPanel OptionEnumerationControl Control for choosing enumeration value OptionFontControl Control for choosing font option OptionsDialogue Options dialogue Op
19. Inselberg Alfred 2010b Parallel Coordinates How it happened http www cs tau ac il aiisreal Cited on page 13 Inselberg Alfred and Bernard Dimsdale 1990 Parallel Coordinates A Tool for Visualizing Multi Dimensional Geometry In Proc 1 IEEE Conference on Visualization Vis90 pages 361 378 IEEE Computer Society ISBN 0818620838 doi 10 1109 VISUAL 1990 146402 Cited on page 13 International Organization for Standardization 2003 The C Standard Incorporating Technical Corrigendum No 1 Second Edition Wiley ISBN 0470846747 Cited on page 53 International Organization for Standardization 2010 Working Draft Standard for Programming Lan guage C http www open std org jtcl sc22 wg21 docs papers 2010 n3225 paf Cited on page 53 Jackson Michael 2008 RGB to HSL and RGB to HSV Color Model Con version Algorithms in JavaScript http mjijackson com 2008 02 rgb to hsl and rgb to hsv color model conversion algorithms in javascript Cited on page 71 Jern Mikael 2009 Collaborative Web Enabled GeoAnalytics Applied to OECD Regional Data In Proc 6 International Conference on Cooperative Design Visualization and Engineering CDVE 09 Lecture Notes in Computer Science volume 5738 pages 32 43 Springer doi 10 1007 978 3 642 04265 2_5 Cited on page 46 124 Bibliography Jing Wang Peng Wei Matthew O Ward and Elke A Rundensteiner 2003 Interactive Hierarchical Dimension
20. Mapping from a font name to a font file s path is a very complex task as well The thesis explains how these problems were solved The conclusion of the thesis in Chapter 6 is about the future of parallel coordinates and how the development of MDE might continue Parallel coordinates are evolving in many different directions and the problem of too many lines cluttering the display has not been completely solved yet Both the theory of parallel coordinates and their practical implementation are still works in progress More work by software engineers and usability engineers are needed to support their use in actual data analysis Appendix A is MDE s user manual and explains the features of MDE with practical examples and many supporting screenshots Appendix B on the other hand is a detailed guide for developers adding new visualisations to MDE Chapter 2 Information Visualisation When I tell you Observe please then you should according to your typical language use ask me Fine but what What is it that you want me to observe In other words you will ask me to specify a problem which can be solved by your observation Sir Karl Popper Alles Leben ist Probleml sen 1996 pp 19 20 Information visualisation is defined by Andrews 2010 as the visual presentation of abstract infor mation spaces and structures to facilitate their rapid assimilation and understanding In other words the goal of informa
21. UBS 2009 using a basic scatter plot display implemented in MD VS a software tool developed by Mader 2007 which is discussed in detail in Chapter 3 Ahlberg and Shneiderman 1994 provide an example of the scatter plot technique applied to the task of finding movies in software called FilmFinder which is shown in Figure 2 6 The horizontal and vertical axes of the plot displayed by the software correspond to popularity and age while the colour of the plotted points is used as an additional channel conveying information about a film s category 2 3 2 Histograms Histograms are a common information visualisation technique invented by Pearson 1895 In their most basic form they map one dimensional data on a two dimensional visualisation The distribution of data points on the horizontal axis is projected on the vertical axis or vice versa as shown in Figure 2 7 In order to do so the range of the first axis is divided into subranges of equal length so called bins The size of the bins in the other dimension is determined by the number of data points belonging to the corresponding subrange Wand 1997 explains mathematical approaches to determine appropriate bin width or number of bins 2 3 3 Star Plots Star plots Chambers et al 1983 are a technique for visualisation of multi dimensional data Friendly 1991 describes star plots as a series of depictions for each record in a data set Each of those smaller depictions resemble
22. 1 31 32 82 12 64 73 records 73 used 0 selected 52 dimensions Figure 3 3 Parallel coordinates visualisation with Multi Dimensional Explorer MDE the soft ware developed for this thesis In this example the application visualises the data set about prices and earnings in different cities around the globe UBS 2009 turned into a comma separated value file suitable for opening in MDE with the InfoScope tool made by Brodbeck and Girardin 2003 73 records and 24 dimensions are shown at the same time Every polyline corresponds to one record The record for the city of Milan and all its data points have been highlighted in red 15 3 Parallel Coordinates hyg small csv Multi Dimensional Explorer Parallel Coordinates 8 xj w File Views Window Help 8 x E 2 Parallel coordinates scatter Plot Records Dimensions 4 8672 12391 236890 9110 GJ1005A 101Psc 268G Cet 2 6595689 89 264138 10000000 0 All 1 ZetTuc pEridani 0 00025315 88 362875 3 567988 0 45 9 62 O9llinn StarlD Hip HD HR Gliese BayerFlamsteed ProperName RA Dec Distance Mag AbsMag Spectrum Colorin SVG Export Zoom 100 polyline Opacity 100 J T Rotatetabes 50 _T Meantine 7 Polyines Histograms I Left I Right Histogram Bins 5 f Options 3000 records 3000 used 0 selected 14 dimensions Figure 3 4 Too many polylines can easily clutter a parallel coordinates visualisation This ex ample
23. 2 Class Matrix stores std vector of Point for representing matrix fields A union member variable is used in Point to store for each field either a double value for real number dimensions or an int value for integer number and enumeration dimensions In the latter case the value is the index of the enumeration value Information pertaining to entire dimensions is stored by Matrix in a separate collection of Dimension instances Holder is for most of MDE s other source code sections the only way to communicate with Matrix It permits only non modifying operations on it providing a const reference to Matrix Among those are pointString and pointNormalised used to obtain a string representation of a field or its normali sation in the range 0 0 1 0 irrespective of the corresponding dimension s type Figure 4 3 explains the difference between obtaining string representations and normalised values and how those operations are related to the way Mat rix stores and structures data Modifying operations of Matrix are mirrored in the interface of Holder which intercepts the calls and notifies any registered Observer instances This is a classical application of the Observer design pattern by Gamma et al 1994 After sending the notification Holder delegates the operation to Matrix This is similar to the Proxy design pattern by Gamma et al 1994 but in the case of MDE the approach 1s not object oriented because there is no run time polymorp
24. Cited on pages 8 12 and 18 Wand M P 1997 Data Based Choice of Histogram Bin Width The American Statistician 51 1 pages 59 64 ISSN 0003 1305 http www jstor org stable 2684697 Cited on page 8 Ward Matthew O 1994 XimdvTool Integrating Multiple Methods for Visualizing Multivariate Data In Proc 5 IEEE Conference on Visualization Vis 94 pages 326 333 doi 10 1109 VISUAL 1994 346302 http davis wpi edu xmdv docs vis94 paf Cited on page 26 Wegenkittl Rainer Helwig L ffelmann and Eduard Gr ller 1997 Visualizing the Behavior of Higher Dimensional Dynamical Systems In Proc 7 IEEE Conference on Visualization Vis97 pages 119 126 IEEE Computer Society ISBN 0818682620 doi 10 1109 VISUAL 1997 663867 http www cg tuwien ac at research vis dynsys ndim ndim_crc paf Cited on page 22 Wegman Edward J and Qiang Luo 1996 High Dimensional Clustering Using Paral lel Coordinates and the Grand Tour Computing Science and Statistics 28 pages 361 368 http courses cit cornell edu eceprojectsland STUDENTPROJ 2007t02008 ak364 491_ak364 tecrep_pdf pdf Cited on page 17 Wellcome Library 2011a Diagram of the Causes of Mortality in the Army in the East by Florence Nightingale http images wellcome ac uk indexplus image L0041105 html Cited on page 4 Wellcome Library 2011b Map showing deaths from Cholera in Broad Street Golden Square and the neighbourhood by John Snow http ima
25. Internet a web address is given in its bibliography entry 3 5 1 MDVS The Multi Dimensional Visualisation System MDVS was made by Hackl 2010 as an extension of the Java Oracle 2011 desktop application MultiDimensional MetaData Visualization System MD7VS 32 hyg small csv Multi Dimensional Explorer Parallel Coordinates 3 Parallel Coordinates wh File Views Window Help E Parallel coordinates scatter Plot Records Dimensions ES 181xi la x 4 8672 t 12391 4 236890 4 9110 4 GJ1005A 101Psc 268G Cet 2 6595689 89 264138 10000000 0 E 1 2 1 2 Wo9848 ZetTuc pEridani 0 00025315 88 362875 3 567988 0 45 9 62 O9llinn 0 212 StarlD Hip HD HR Gliese BayerFlamsteed ProperName RA Dec Distance Mag AbsMag Spectrum Colorin SVG Export Zoom 1001 Polyine Opacity 50 H T Rotate Labels 30 H I MeanLine I Polyines Histograms I Left I Right Histogram Bins 5 Options 3000 records 3000 used O selected 14 dimensions a Before focussing hyg small csv Multi Dimensional Explorer Parallel Coordinates wh File Views Window Help E 2 Parallel coordinates scatter Plot Records Dimensions E laxi 181x 4 4 4 t 8672 12391 236890 9110 236890 4 GJ1005A 101Psc 268G Cet 2 6595689 89 264138 10000000 0 mn 1 2 1 2 Wo9848 ZetTuc pEridani 0 00025315 88 362875 3 567988 0 45 9 62 O9ilinn 0 212
26. Modules i ecreis aa ara ed ba a a be ee hw ee 55 4 2 MDE Application Components 2 2 CC mn 58 4 3 MDE Data Components 62 4 4 MDE Data Input Components ooa 64 4 5 MDE s Options Components nn 66 4 6 General Components in MDE s Presentation Module 72 4 7 Specific Components in MDE s Presentation Module 73 4 8 MDE s Presentation Structure Components 2 200 75 4 9 Unit Tests for MDE s Data Module o e 77 4 10 Unit Tests for MDE s Data Input Module o 78 4 11 Other MDE Unit Tests 2 2 0 0000 20 a 79 vi List of Listings B 1 New MDE Visualisation Header File 2 Coon ee ee 112 B 2 New MDE Visualisation Source File 2 2 ee 113 B 3 New MDE Visualisation Factory Header File o o 114 B 4 New MDE Visualisation Factory Source File o o e 115 B 5 DataViewFactoryGroup s Constructor in MDE o o 115 B 6 MDE TestVisualisation Implementation o o e e 117 vii Vili Acknowledgements First and foremost thanks go to my parents for their support throughout the years of my studies Thanks also go to my older brother Bernhard who once got me into software engineering by encouraging me to learn BASIC V2 on our Commodore 64 at the age of 12 Some years later he would a
27. Proc 10 IEEE Symposium on Informa tion Visualization InfoVis 2004 pages 81 88 IEEE Computer Society doi 10 1109 INFVIS 2004 68 Cited on page 17 Baldonado Michelle Q Wang Allison Woodruff and Allan Kuchinsky 2000 Guidelines for Using Multiple Views in Information Visualization In Proc 4 Working Conference on Advanced Visual Interfaces AVI 2000 doi 10 1145 345513 345271 Cited on page 7 Bauer Alois Sophie Steinparz Richard A mair and John Feiner 2010 Parallel Coordinates A Tutorial http courses iicm tugraz at ivis Cited on pages 22 and 30 Bendix Fabian Robert Kosara and Helwig Hauser 2005 Parallel Sets Visual Analysis of Categorical Data In Proc 11 IEEE Symposium on Information Visualization InfoVis 2005 pages 133 140 IEEE Computer Society ISBN 078039464x doi 10 1109 INFVIS 2005 1532139 http www bendix at parallel sets infovis 2005 bendix pdf Cited on pages 19 and 20 Bertini Enrico Luigi Dell Aquila and Giuseppe Santucci 2005 SpringView Cooperation of Radviz and Parallel Coordinates for View Optimization and Clutter Reduction In Proc 3 International Conference on Coordinated amp Multiple Views in Exploratory Visualization CMV 2005 pages 22 29 IEEE Computer Society ISBN 078039464x doi 10 1109 CMV 2005 17 http www dis uniromal it santucci VisDis Papers CMV05 pdf Cited on page 7 Boost 2010 Boost C Libraries http www boost org Cited on pag
28. Situvis Situvis is an open source parallel coordinates visualisation software package developed by Clear et al 2009 within the Processing programming environment Fry and Reas 2010 The tool is meant to be used for modelling behaviour patterns and for conflict analysis although it could also be used more generally for all kinds of input data It operates on text based t sv files Originally before the software was labelled Situvis its most particular feature was to pair parallel coordinates with a node like graph visualisation of the same data set Figure 3 23 shows a comparison of the current software with its predecessor which was given the name paired parallel coordinates It was based on the observation by Shannon et al 2008 that parallel coordinates are not particularly good at visualising relationships between records The graph visualisation component however has been removed and is no longer present in the current Situvis software User interaction in Situvis is relatively limited Axes can be reordered but an axis cannot be removed or repeated When the user moves the mouse pointer over an axis a tooltip shows the value corresponding to the mouse pointer location Records can be selected by intersecting their polylines with a line the user draws by dragging the mouse over the visualisation The records actual values however are not displayed There is no table view of the data either Apart from the line int
29. accordance with C rules names of classes and free standing functions in this section are not written with the full namespace prefix For example mde presentation DataView is Written as DataView in the Data subsection and as presentation Data View in the other subsections 4 2 1 Application This module contains components which are not of general use but belong to the concrete MDE GUI application specific windows menus bars and dialogues Its central class is Mainwindow which repre sents the main window of MDE the parent of any other windows which might be opened The module depends on all of the Data Data Input Presentation and Options modules in the following way e The data storage and synchronisation mechanisms of Data are used to provide a central point of reference to the data set which is accessed by visualisation subwindows e Data Input is used to link file open dialogues to actual file parsing mechanisms e Presentation provides the contents of visualisation subwindows in the GUI e General program options are defined and maintained via the Options module wx Widgets 2010 is the GUI framework used by MDE It provides an abstract interface equal across different platforms internally implementing it by direct access to the operating system s Application Programming Interface API This aids in achieving platform specific look and feel increasing usability of a software application Much startup initialisati
30. animation hierarchical clustering VisDB Offers other visualisations Caleydo Aimed at medical science OECD eXplorer OECD specific analyses direct online use Protovis Implemented entirely in JavaScript and SVG VizCraft Aimed at Aircraft Design Liquid Diagrams Integration into Google Docs SVG export Table 3 2 A summary of parallel coordinates software reviewed in this thesis For each of the tools main characteristics are named previously developed by Mader 2007 The original MD VS application lacked a parallel coordinates visualisation It had however been designed with potential extensions for further visualisation schemes in mind MDVS therefore integrated a parallel coordinates component Figure 3 19 is an example of a test data set visualised with parallel coordinates in MDVS The component offers the following user interaction features Filtering In order to render a record unused the user has to restrict the dimension by dragging the blue triangles located on the top and bottom of the corresponding axes All records whose lines do not pass through the now restricted vertical area will be considered unused and appear greyed out Record Selection Records which are not unused can be selected and deselected either by click ing the corresponding lines or by drawing a selection rectangle with the mouse Selected records appear in blue unselected ones in black Figure 3 19 shows the difference between unused unse l
31. be used to identify all the available font names and their correspond ing file names because the relationships are stored there at key HKEY_LOCAL_MACHINE SOFTWARE Microsoft WindowsNT CurrentVersion Fonts The key is opened with RegOpenkeyEx fol lowed by a series of calls to RegEnumValue and RegQueryValueEx to retrieve font names and file names Any entry whose name does not end with TrueType or whose filename does not end with ttf is excluded from the results because this is the easiest way to retrieve only TrueType fonts Clients of getAvailableFonts and getFontDirectory are oblivious to these system specific low level intricacies They operate instead on safe C constructs like std map std string and 88 5 Selected Details of the Implementation boost filesystem path Consequently porting the mechanism to other platforms requires no mod ification of client code 5 2 2 Drawing Text Having obtained a map of all available fonts with corresponding file paths FTGL can be used to draw text respecting the user s font choice All use of FTGL in MDE just as any OpenGL specific MDE functionality in general is encapsulated in class visualisationImplementationopenGL Two different classes for FTGL font rendering are used in the two overloaded versions of doDrawText FTGLTextureFont is used in the overload taking a rotation parameter for drawing potentially rotated text
32. black to red relative to their value in the corresponding dimension Selecting records or obtaining detailed information about the records in general is entirely absent from the software s interface 3 5 14 Liquid Diagrams Liquid Diagrams provide parallel coordinates and a number of other visualisation techniques as a Google Docs Gadget Google 2011 They were implemented by Andrews and Lessacher 2010 The idea is that a user enters data into a Google Docs spreadsheet and then inserts the gadget to visualise it as shown in Figure 3 32 All of this happens online with no installation of additional software required The parallel coordinates component of Liquid Diagrams features extensive support for user inter action Axes can be dragged horizontally with the mouse to reorder dimensions An axis can also be flipped by clicking on a small arrow above it Polylines are selected by simply clicking on them Furthermore there are triangular handles on the top and bottom of each axis which can be dragged vertically by the user to define a range on the corresponding dimension all polylines not going through that range are greyed out and cannot be selected as a minor problem the tool does not make sure that greyed out lines are drawn below the others 3 5 Software Applications Figure 3 31 Parallel coordinates in VizCraft Goel 1999 a visualisation tool written in the con text of aircraft design implemented as a Java
33. con siderations have been taken into account for efficient mass storage of records This is the reason class Point uses a union To mitigate security problems related to the use of unions in C access to the data inside of Point is restricted exclusively to Matrix Table 4 3 gives an overview of all classes free standing functions enums and typedefs found in the Data module 60 Observer Stores an instance Maintains filters std vector lt Point gt CA al v Dimension Information pertaining to whole dimension Name Visibility Type Bounds Enumeration Values Observer Holder Matrix 4 Multi Dimensional Explorer Observer Non const operations of Matrix are mirrored in the interface of Holder Matrix modifications are intercepted Other Observers are notified Holder provides Observers with Matrix const pointNormalised pointString Point union Value int int value double double value y Information pertaining to single field int for dimensions of type Integer or Enumeration as enumeration index double for dimensions of type Real Figure 4 2 Multi dimensional data storage and synchronisation in MDE s Data module Inter nally class Matrix stores a std vector lt Point gt Point represents a field and stores either a double value for real number dimensions or an int value for integer number and enumera tion dimen
34. de a 37 Parallax cr soca acaso wee ba ee wR eka eS Se we oe ae 38 DILUVIS Sade Gow ha a RA Saw rs el aed bP a AA bw a 41 PICVIZ a 2 db Be ae GE ae bed Oe BH hae a ee ead Se 42 lil 329 OO ee ea we ek ee eo ee ee On ea ee ee ot 43 3 26 XmdyTool u 20 eh ae eA BA A OS ba dd ee ee eae nd 44 3 27 VDB 4 4 8 4 3 ton a we heh Se EDR ba ed bok ed eed 45 3 28 Caleydo a er a sas a ee aa a A ele a ie 46 3 29 OECD explorer eii ee ara a a ae ee tw a a 47 3 30 PLOTOV S vaca 450 aac sh en un a Ae ae ede Edw pl aca ee bdo age eo aw a a 48 SOU VIZA o eas a ached GSE Stee amp ea ae he a ele SS Sate eS 49 3 32 Liquid Diagrams estoi 0 aan ee A an ee 50 4 1 MDE Module Dependencies 0 0 200000000000 Fee 56 4 2 MDE Storing and Synchronising Multi Dimensional Data 60 4 3 Structure of an MDE Matrix 2 2 ee 61 4 4 MDE s CSV Support Architecture 2 ee ee 63 4 5 Options Example in MDE e 65 4 6 MDE s Concrete DataView Subclasses o o e e 68 4 7 MDE s Visualisation Class Hierarchy o nn 69 4 8 MDE s Visualisation Frontend and Backend o o o 70 4 9 MDE s Visualisation Structuring e 74 4 10 MDE Application Window nn 76 5 1 Multiple Visualisation Steps in MDE o 00000200004 85 5 2 Font Names and Files ee ee ee ea a ee a ee 87 9 3 Text Rotation 2 424 nt ed be Poe ee ee e
35. dialogue window which appears Due to wxWidgets reliance on platform specific components in its implementation it is imperative for this new source file to enable Microsoft s language extensions which are disabled by default in all files This can be done on the file s Language property page Failure to do so will result in many seemingly unrelated compiler errors Listing B 1 is the header file for TestVisualisation Listing B 2 is its source file Note how the only requirement for the class is its constructor taking a DataViewConstructionParameters argu ment because this is passed to the base constructor and to implement the pure virtual private function partiallyCreateVisualisation 112 B MDE Developer Guide Fie 74 Solution mde 7 projects 1 mde application TE mde data TE mde data input 5 mde options A mde presentation fll External Dependencies BE Headerdatcjags al Quelldateies Add F 1 AE a mde presentatic Cut Ctrl x Shift Alt A 3 mde test a 3 Copy Ctrl C 7 Paste Ctrl V X Delete Del Rename F2 Figure B 3 The first step in adding a new visualisation to MDE is to create its header file in the Solution Explorer of Visual C 1 ifndef MDE PRESENTATION TEST VISUALISATION H 2 define MDE PRESENTATION TEST VISUALISATION _H_ 3 4 include visualisation h
36. e x Fie View Brushes Tools Backend Help ea O xz00m fo a vzcom as x a 50 x I ft_police unemp manu_wrkrs handqun_Ies gov_wrkrs cleared homicides 621 40 1179 95 35 75 96 18 54 52 396 68 11 39 51 de E e at H 253 86 2 81 447 60 107 67 29 05 57 13 a a manu_wkrs 492 89 4 gt Figure 3 26 XmdvTool s parallel coordinates component XmdvTool is an open source C ap plication suitable for general purpose visualisation of multivariate data The interior of the grey polygon behind the lines defines an interval on each axis and thus the current record selection XmdvTool also features clustering hierarchical parallel coordinates for large data sets and animation for data sets in which one dimension represents time e The data set can be sampled subsetting All of these manipulations immediately update the parallel coordinates display and all other visuali sations 3 5 8 XmdvTool XmdvTool is an open source application for general purpose data visualisation implemented in C using OpenGL Shreiner 2009 Khronos Group 201 1a and Tcl Tk 2010 It features scatter plots star glyphs dimensional stacking and parallel coordinates For each of its visualisation schemes XmdvTool also features a hierarchical view that is clustering Rundensteiner et al 2007 XmdvTool 2010 Xmd vTool uses text based okc files for dat
37. file cannot be opened for reading CsvFileReader Creates a Matrix froma CSV file CsvParsingOptions Options for CSV parsing FileReader Creates a Matrix from a file FileReaderGroup FileReader instances with associated file name endings IsDefaultValidLine Text StreamInput s default functor to determine whether to skip lines Text StreamInput Provides Text StreamInput Iterator pair for Matrix creation from a std istream Text StreamInput Iterator Iterates over a std ist ream and turns each line into a Text StreamInputLine Text StreamInputLine Tokenises a line of text for Mat rix input TooFewColumnsInCsvLineException Exception when a CSV line contains too few columns TooManyColumnsInCsvLineException Exception when a CSV line contains too many columns guessCsvParsingOptions Guesses CsvParsingOptions suitable for a specific CSV file makeDimension Creates a Dimension by analysing text input makeDimensionImplementation Implementation of makeDimension Table 4 4 The classes and free standing functions of MDE s Data Input module modules Options are persistently stored in an Extensible Markup Language XML file Bray et al 2008 and the TinyXML library by Thomason et al 2010 is used for parsing it No library is used for writing XML because the process of writing XML is too simple to justify adding yet another 3 party library to MDE The module is built around a tree for cascading option values
38. in Multi Dimensional Explorer MDE shows the result of rendering 3000 polylines from a data set of 3000 records and 14 dimensions a subset of a data set about stars Nash 2006 The massive number of lines clutter the display and make the visualisation unusable various techniques to reduce clutter have been proposed hyg small csv Multi Dimensional Explorer Parallel Coordinates laj x w Fie Views Window Help l8j xj E 2 _Paralle Coordinates Scatter Plot Records Dimensions 8672 12391 236890 9110 GJ1005A 101Psc 268G Cet 2 6595689 89 264138 10000000 0 13 06 14 199165 2 835 1 2 1 2 Wo9848 ZetTuc pEridani 0 00025315 88 362875 3 567988 0 45 9 62 O9llinn 0 212 StarlD Hip HD HR Gliese BayerFlamsteed ProperName RA Dec Distance Mag AbsMag Spectrum Colorin SVG Export Zoom 100 Polyine Opacity 1 _ PT Rotate Labels 53 H I Meantine I7 Polyines Histograms I Left I Right Histogram Bins 5 Options 3000 records 3000 used 0 selected 14 dimensions Figure 3 5 When the sheer number of polylines clutter the display of a parallel coordinates visual isation reduced opacity can mitigate the problem This is another example of Multi Dimensional Explorer MDE showing the same data set as Figure 3 4 The opacity of each polyline however has been reduced to 1 so the user can once again recognise certain patterns in the data 3 2 Handling Large Data Sets 17 3 2 H
39. it This is an established user interface paradigm 3 4 Using Parallel Coordinates cities2009 csv Multi Dimensional Explorer Parallel Coordinates SuGExport zoom 100 f Polyine Opacty 50 H f BotateLabels 0 2 I Meantine I Polyines Histograms I Left I Right Histogram Bins 5 Options l 73 records 73 used 0 selected 52 dimensions a A parallel coordinates axis is being moved shown in red here during movement cities2009 csv Multi Dimensional Explorer Parallel Coordinates 218 xi SvGExport zooms 100 Polyine Opacty 50 j M kaiii 30 J Meantine Y Polyines Histograms I Left I Right Histogram Bins 5 f Options 73 records 73 used 0 selected 52 dimensions b After moving dimensions about vacation and rent are located next to each other Figure 3 12 In parallel coordinates users sometimes want to move axes so that dimensions whose correlations they wish to explore are placed next to each other For example the user might want to study the correlation between Medium Rents and Paid Vacation Days Per Year a shows the axis about medium rents being moved by dragging it with the mouse b shows the result of the operation the two axes are located next to each other 27 3 Parallel Coordinates cities2009 csv Multi Dimensional Explorer Parallel Coordinates tn
40. made known to the rest of MDE via DataViewFactoryGroup e ParallelCoordinatesVisualisation derived from InteractiveVisualisation visualises data with parallel coordinates see Chapter 3 ScatterPlotVisualisation derived from InteractiveVisualisation visualises data with a Scatter Plot see Chapter 2 RecordsTableDataView derived from DataView shows individual records in a tabular form DimensionTableDataView derived from DataView gives an overview of the dimensions in the data set Figure 4 6 shows how these concrete subclasses are presented to the user in MDE s GUI Figure 4 7 is a class diagram showing the inheritance relationships Visualisation works with normalised window coordinates in the range 0 0 1 0 The point of ori gin 0 0 0 0 is the lower left corner of a visualisation window This means subclasses do not have to cope with window size and that their visualisations are automatically scalable In some cases however it is still desirable to translate relative to absolute coordinates and vice versa for example when deter mining whether the mouse pointer is close enough to some element of the visualisation or to implement minimum pixel sizes for otherwise freely stretchable elements To fulfil those needs visualisation offers the methods toPixelsX toPixelsY toRelativeX and toRelativeY Subclasses of Visualisation directly or indirectly via InteractiveVisualisation suchas Parallel Coordinates
41. nurse Florence Nightingale shown in Figure 2 1 can be cited as another historical example of creating abstract graphical representations of information long before computers came into existence In 1857 Nightingale created a chart representation of statistical numbers of deaths during the Crimean Wars proving that more soldiers fell victim to poor sanitary conditions in the armies rather than being killed by the enemy Wielemaker 2010 Today Nightingale s chart is often referred to as a Coxcomb Chart although according to Small 1998 she never used that name herself Ancient maps are often cited as historical examples for information visualisation Applying the definition from this chapter s introductory paragraph however they usually cannot be considered as such because the data visualised by them is not abstract they are just visualisations of the real physical positions of cities countries or other places Therefore in such cases it is more correct to talk about geographical visualisation Andrews 2010 names John Snow s historical cholera map from 1854 shown in Figure 2 2 as a famous example of geographical visualisation not to be considered information visualisation because it is primarily based on topographical data The map visualised deaths caused by cholera in different parts of London illustrating Snow s theory of the outbreak of the epidemic in the city to be caused by a specific water pump However contrary t
42. parallel coordinates visualisation displaying many di mensions is to hide some of the axes This can be done in MDE by drawing a selection rectangle with the mouse in the display area above or below the polylines Axes under the rectangle are selected and are displayed in a different colour as shown in Figure A 4 It is now possible to right click on any of the selected dimensions to open a context menu with a Hide item This is how a group of axes can be hidden from the view at once Figure A 5 shows what the user can do with each axis of the parallel coordinates visualisation An axis can be flipped by clicking on the arrow symbol on top of each axis as shown in Figure A 6 Filters can be defined on dimensions by dragging the triangles on the axis line this results in polylines outside of the range becoming unused and thus displayed in grey Furthermore the entire axis can be moved horizontally by grabbing one of its top or bottom elements with the mouse and dragging it around This is so that the user can change the order in which dimensions appear in the parallel coordinates visualisation It is also possible to define more than one filter for an axis Filters can be fine tuned by right clicking on one of their handles and choosing Edit Filter An additional filter type exists for Enumeration dimensions such as the Cities dimension of the Cities data set which contains the names of cities Filters for such dimensions can def
43. set of prices and earnings in different cities around the globe provided by UBS 2009 and turned into a comma separated value file suitable for opening in MDE with InfoScope Brodbeck and Girardin 2003 The 6 bin histogram shows distribution of records in the Gross purchasing power dimension Each bin represents a subrange of the dimension s entire range which spans from 12 0 to 115 8 The wider a bin the more records belong to the corresponding subrange 11 12 2 Information Visualisation eal view O El W Fle View Brushes Tools Window Help la x A mif zoomy 100 zoomdrectonsfx z OR Gy CR OO ix h f Ab y A 4 Glyph 76 Cal 110 00 Prot 2 00 Fat 1 00 Sod 70 00 F ber 1 00 Carbs 3 00 Sugar 15 00 Potas 40 00 V M 25 00 Wt_sv 1 00 Cp_sv 0 75 Figure 2 8 Star plot visualisation of the cereals data set provided by Velleman 1996 in Xmd vTool 2010 Each record in the data set is visualised by one of the stars each ray of a star corresponding to one of the dimensions and each ray s length to the value of the record in that dimension relative to other records In this example the stars are sorted by the Sugar dimen sion so it is easy to see how the Sugar ray the 8 in clockwise direction here becomes longer as the Sugar value becomes greater ioixi Zoom 244 14 W Antialiasing Vectornumbers 7 VectorCluster Figure 2 9 VisIslands displ
44. similar to the logic of Cascading Style Sheets Celik et al 2010 Its central component is class Tree which stores an internal tree of Node objects For every option available in the application a Definition object exists including information about an option s name description type and default value A collection of Definition objects is contained within each Node All definitions are automatically inherited In addition to that each Node also contains values for definitions Values are simply strings whether their contents are valid or not depends on the Definition they are assigned to Every Node must contain a value for each of its own definitions and may contain a value for each of its inherited definitions This is how values are overridden Node s own and inherited values can be read and modified uniformly The class will recognise if the definition of a particular value is inherited or not In order to prevent inheritance of unused options or inheritance of options which would be mean ingless for a subnode imagine a Rendering Engine option in a base node which is not used by a subnode representing a plain text panel inherited options have to be made visible explicitly by calling setVisibleInheritedOptions Figure 4 5 gives a fictional example of an options tree whose root element contains the two general options background_colour and label_font Furthermore it contains two subnodes My View and Your View Your View con
45. std string 2 Compliance with the C language standard to ease porting MDE to other platforms no use of proprietary compiler extensions 3 Heavy use of the Non Virtual Interface NVI idiom Sutter 2001 Public member functions are preferably non virtual delegating to virtual or pure virtual private member functions Protected member functions are preferably non virtual as well 4 Liberal use of the assert macro for safe termination of the application in face of programing errors rather than throwing exceptions Exceptions are reserved for exceptional problems not caused by errors in MDE s source code and in certain cases for a simple means of reporting errors from recursive functions In other cases function return values are deemed appropriate Assertions are not turned off in release builds of MDE this is on purpose 5 Restriction of object oriented features virtual functions to cases where run time polymorphism is actually needed Specifically mde presentation structure VisualisationElement and mde data Point are examples of two classes whose instances can semantically belong to dif ferent types but this typing is not expressed in terms of derivation and overriding virtual functions C was chosen for this project for a number of reasons Its expressiveness and abstraction features ease the creation of complex software applications Unlike other high level languages however C does not impose any cost in per
46. the points In order to aggregate selections a checkbox labelled consistent has to be checked while records are selected On the positive side the application gives users complete control over the way a selection is displayed Colour style and thickness of a selected record s lines can be chosen freely Additionally dimension axes can be reordered and removed from the display They cannot however be repeated In order to identify individual records the user has to activate the Identify interaction mode and move the mouse pointer over the record s lines The record will be focussed and tooltips showing the record s actual data values will appear A particular feature of GGobi s parallel coordinates is to change the visualisation s entire orientation so that axes appear as horizontal lines as opposed to the traditional way of displaying them vertically The application does not feature anti aliasing or zooming GGobi s most distinguishing feature is not directly related to its parallel coordinates visualisation component It is the tool s extensive set of data manipulation and statistical tools Examples of such tools include the following Swayne et al 2006 Data in each dimension can be transformed standardised and sorted e Random noise can be added e Records can be selected automatically according to defined rules automatic brushing 44 3 Parallel Coordinates XmdvTool 7 0
47. traditionally use straight lines to connect axes This leads to the problem of polylines crossing the same point making it impossible to tell which lines belong to which polylines Graham and Kennedy 2003 call this the cross over problem and suggest the use of curves as a solution to overcome it Using curves as shown in Figure 3 9 allows users to immediately keep apart entire polylines using natural visual perception even when all polylines cross at the same point on the same axis The paper argues that except for their end points the lines in traditional parallel coordinates visualisations carry no information of their own so other means of connecting points on the axes can be employed without any loss of information This is in line with Theisel 2000 who notes that the horizontal space between parallel coordinates axes can be leveraged to convey more information 3 3 4 Continuous Parallel Coordinates Continuous parallel coordinates were introduced by Heinrich and Weiskopf 2009 They argue that tra ditional parallel coordinates techniques miss an important aspect of the input data the continuous nature of data gathered from scientific observations Discretisation of such data in visualisations inevitably leads to a loss of information conveyed Therefore mathematical techniques to express the continuity of input values in an adapted form of parallel coordinates have been devised 3 4 Using Parallel Coordinates This section discu
48. und OpenGL entwickelt Sie erleichtert die empirische Datenanalyse durch hohe Performance und eine Reihe von Interaktionsm glichkeiten Die Arbeit dokumentiert die modulare Softwarearchitektur von MDE Der Anhang enth lt Anleitungen zum Gebrauch von MDE und zu seiner k nftigen Weiterentwicklung Statutory Declaration I declare that I have authored this thesis independently that I have not used other than the declared sources resources and that I have explicitly marked all material which has been quoted either literally or by content from the used sources Place Date Signature Eidesstattliche Erklarung Ich erkl re an Eides statt dass ich die vorliegende Arbeit selbst ndig verfasst andere als die angegebe nen Quellen Hilfsmittel nicht benutzt und die den benutzten Quellen w rtlich und inhaltlich entnomme nen Stellen als solche kenntlich gemacht habe Ort Datum Unterschrift Contents Contents List of Figures List of Tables List of Listings Acknowledgements Credits 1 Introduction 2 Information Visualisation 2 1 o e ice ga Beh ea BR Bo hk a es ac ee Ree ho SD Ane es 2 2 Interaction z a 8 4 a 4 8626 bb boobed BAG La wa hacked eee ed bas 2 3 Multi Dimensional Data Visualisation 2 2 2 2 2 oo mn nn 3 Parallel Coordinates 3 1 Backgiound ck u u u u ua Be A e ne a aiia 3 2 Handling Large Data Sets 2 2 aaa 3 3 Varlations g sauda a a ea Be ee eae a 3 4 Using Parallel Coordi
49. 0 IV Visible Filters IV Use filters IE a Dimensions Window 4 2 2 4 4 9370 93700 b Axis Before Change c Axis After Change Figure A 15 MDE s dimensions view shown in a is used to fine tune information about entire dimensions In this example the user modified the minimum and maximum value of the Vaca tion dimension This is all synchronised with the parallel coordinates visualisation Before the modification the dimension ranged from 6 to 30 as shown in b After the change as shown in c it ranges from 0 to 100 106 A MDE User Guide Many options are inherited in this hierarchy and can optionally be overridden For example the general colour to use for selected records can either be set globally on the General page or individually for each category of views All option definitions and the currently chosen values are stored in options xml In addition every instance of a view that is every single subwindow has its own set of inherited options for temporary modification of options If for example a user opens a parallel coordinates visualisation and wants selected records to be displayed in green then there are three ways to perform this task 1 Open the options dialogue and change the Colour Of Selected Records option on the General page This will lead not only to all parallel coordinates windows displaying green polylines but also all scatter plot windows displaying green dots and all re
50. 1 is stored in the last field leading to a string representation of unknown and a normalisation value of 0 5 61 62 4 Multi Dimensional Explorer Name Purpose Change Records data manipulations made via Holder Dimension Represents a data dimension DimensionFilter Restriction of records to a subset in one of the dimensions DimensionIndex Index of a dimension in Matrix DimensionPosition Position of a dimension in Matrix when dimensions are rearranged Exception Base exception for data related errors Holder Filters and synchronisation for Matrix HolderBatchCommit Commits multiple data changes in one batch IllegalInputRowLengthException Exception when an input row has too few or too many fields Observer Observes Holder Point Field of Matrix PointConversionException Exception when Matrix construction fails due to conver sion error PointStringConverter Converts strings to Point instances RecordState State of a record in Matrix addDefaultFiltersToHolder Adds default limitless filters to Holder dimensionFilterDescription Textual representation of DimensionFilter getUsedRecordIndices Applies DimensionFilter range to a dimension obtaining records not excluded through the filters prettyReal Pretty representation of real number typeDescription Text description of a Type
51. 16000 4 4 displayed weight IT I 33 397 selected tuple u m 1 E i Er X coordinate attribute 0 c Curve Width 1 ERS exit Y coordinate attribute 1 c redau Figure 3 27 VisDB s parallel coordinates component A data set with three dimensions is visu alised on the left side The right side of the screen contains user interaction elements controlling the number of records displayed as well as the vertical stretching of each axis at the same time and combined with logical operators AND OR and NOT in order to create complex selections Time series animation is a special feature of XmdvTool If one of the data set s dimensions can be considered to represent time then the data set can be animated as a series of visualisations changing over time XmdvTool allows zooming both horizontally and vertically Despite the application using OpenGL the multiple visualisations it produces are not anti aliased Another downside of the software is that although record selections are maintained across different visualisation schemes the multiple visualisa tions cannot be shown together at the same time 3 5 9 VisDB VisDB is a general purpose visualisation tool written in C using Motif 2000 It was developed at the Institute for Computer Science at the University of Munich to support the exploration of large databases Keim and Kriegel 1994 Keim et al 2002 Th
52. 2 a ig Gen ES o ay 7 ong Cy SVG Export zoom 100 f Polyine opacty s 50 H f RotateLabelg 0 J I meantine Y Polyines Histograms I Left I Right Histogram Bins 5 Options I 73 records 73 used 0 selected 52 dimensions Figure 3 13 In parallel coordinates visualisations if many lines cross in the interval between two axes and if the crossings give way to more parallel lines upon flipping either of the axes a correlation between the two dimensions can be deduced In this example which continues the scenario from Figure 3 12 the axis about medium rents has been flipped The left interval marked with a red ellipse still shows many crossed lines so it can be assumed that rents are not much related to the left hand dimension vacation paid working days per year The right interval however marked with an orange ellipse now shows many more parallel lines This indicates a correlation between medium rents and the right hand dimension time required for 1 kg of bread minutes In other words people living in cities with low rents tend to work longer to earn enough money to buy kg of bread Bie Views Eb _Paralei coordinates scatter Piot Records Dimensions Cities Vacation paid working days per year Normal local rent medium USD per month Time required for 1 kg of bread minutes Time re SVG Export Zoom 100 polyine Opacity
53. 26 0 99541884 096273723 0361198518 0 95457247 094112327 094335933 0 8640254 o 033748985 0 000509042 01037386217 0 096224911 0 092851044 0025382179 01145557 0 008026362 0065965903 007174001 0 002732000 untitled untited2 untitled3 unitled4 untiteds untitled united untitled untitleds untitled 10 untile b Main Contents 078840291 0 90966428 098165987 0 99943326 0 99541884 096273723 0936119818 0 95457247 0 94112327 094335933 10 8640254 07282288 GE 02208174 ooadsedact 01037386217 En a Jo caa74aoas 0 000500042 0037388217 0 098224911 0 052951044 9025382179 01145557 0008026362 0065965903 0071746801 0 002732006 untied untitled united unitieda untiteds untitede unites unicas untteds urtitied10 untied c Overlay Figure 5 1 In MDE rendering performance is optimised by breaking it up into three steps In a the background of a parallel coordinates visualisation is drawn this comprises dimension names as well as minimum and maximum labels and flipping arrows on each dimension In b the main contents polylines mean line and filter handles here are drawn on top of the background In c the overlay is drawn by rendering a red highlighted copy of the polyline under the mouse and adding temporary tooltips with the corresponding record s values Unused records are put into the middle layer rather than the background layer due to the observation that the manipulation of filters is a frequent
54. 27448 Cited on page 109 Shadovitz Deborah S 2007 Mac Efficiency 2 How to select items in the Finder http macefficiency com selection finderselection html Cited on page 29 Shannon Ross Thomas Holland and Aaron Quigley 2008 Multivari ate Graph Drawing using Parallel Coordinate Visualisations Technical Re port University College Dublin http www csi ucd ie content multivariate graph drawing using parallel coordinate visualisations Cited on pages xi 7 and 40 Bibliography 127 Shneiderman Ben 1996 The Eyes Have It A Task by Data Type Taxonomy for Information Visualiza tions In Proc 12 IEEE Symposium on Visual Languages VL 96 pages 336 343 IEEE Computer Society ISBN 081867508X doi 10 1109 VL 1996 545307 http www cs uta fi 3t68641 infoviz The_Eyes_Have_It pdf Cited on page 6 Shreiner Dave 2009 OpenGL Programming Guide The Official Guide to Learning OpenGL Versions 3 0 and 3 1 Seventh Edition Addison Wesley Professional ISBN 0321552628 Cited on pages 6 44 46 53 55 68 and 83 Siirtola Harri Tuuli Laivo Tomi Heimonen and Kari Jouko Raiha 2009 Visual Perception of Parallel Coordinate Visualizations In Proc 13 IEEE International Conference on Information Visualisation IVO9 pages 3 9 IEEE Computer Society ISSN 1550 6037 doi 10 1109 IV 2009 25 Cited on page 14 Siirtola Harri and Kari Jouko R ih 2006 Interacting with Parallel Coordinates Inter
55. 4X http isgwww cs uni magdeburg de visual files publications Archive Theisel_2000_VMAV paf Cited on page 22 Thomason Lee Yves Berquin and Andrew Ellerton 2010 TinyXML http www grinninglizard com tinyxmldocs index html Cited on pages 55 64 and 110 Tricaud S bastien and Victor Amaducci 2009 Know Your Tools use Picviz to find attacks https www honeynet org files KYT Picviz_v1_0 paf Cited on page 40 Tricaud S bastien Philippe Saad and Kara Nance 2011 Visualizing Network Activity using Parallel Coordinates In Proc 44th Hawaii International Conference on System Sciences HICSS 44 pages 1 8 IEEE Computer Society ISSN 1530 1605 doi 10 1109 HICSS 2011 488 http wallinfire net picviz Cited on pages 40 and 80 128 Bibliography Tricaud S bastien and Philippe Saad 2010 Applied Parallel Coordinates for Logs and Network Traffic Attack Analysis Journal in Computer Virology 6 pages 1 29 ISSN 1772 9890 doi 10 1007 s11416 009 0127 3 Cited on page 40 UBS 2009 Prices and Earnings http ww ubs com 1 ShowMedia wealthmanagement wealth_management_research prices_earnings contentId 170298 amp name PreiseLoehne_2009_e pdf Cited on pages 8 9 11 12 15 22 24 25 36 and 76 van Heesch Dimitri 2011 Doxygen http www stack nl dimitri doxygen index html Cited on page 54 Velleman Paul 1996 Cereals Datafile http lib stat cmu edu DASL Datafiles Cereals html
56. 5 1 62 3 84 9 Dublin 90 7 106 4 92 7 Frankfurt 92 9 84 6 90 8 Geneva 104 4 100 6 106 8 Helsinki 86 1 88 1 94 5 Hong Kong 41 6 52 3 80 9 Istanbul 29 9 31 0 74 2 Jakarta 12 0 14 5 47 8 Johannesburg 52 8 54 8 48 6 Cairo 22 4 24 1 45 3 Kiev 21 1 23 7 52 1 Copenhagen 115 8 86 2 108 4 Kuala Lumpur 33 1 38 4 43 2 Lima 29 6 32 5 50 5 Lisbon 60 6 66 0 74 0 Ljubljana 68 0 56 4 64 3 London 81 5 86 7 84 6 Table 3 1 Part of the cities data set by UBS 2009 containing information about prices and earnings of various cities in different countries This data set is a popular benchmark for testing information visualisation software There are 73 records one for each city and 52 dimensions including the name of the city 3 4 Using Parallel Coordinates 25 A Fie Ves Whew 2181 x Fer Scatter Plot Records Dimensions Re ee ER ZE NG THTTR ARTA DS TO EEE 419 237 30 0158 84 65 17711 X4 468 780 3118 712 8903 10 640 4 8839 1 24 71830101 1 7 LE apa 30 924 9 5 0 6 1158 6 0 E 0 Yo a Pg ing ingly Mayle hat k fer Srp amp Sy es lotos EIT EN pta oy SER SS OS Pago dut LFI II ER nai ae SEIN 5 DR SIT gt EL Van RER gt RRA Song e ET 5 Lo a pee vn TH oy ag m oop be SO elegy US 0 0007 ones oy EROS fof reo Uso Vo COHEN One see y 7 anes ye mony Cg onta 7 d y Le a hj ony SVGExport zoom 100 4 Polyine opacity 50 H FH I Rotate Labels 20 2
57. 7 explains that for example with multi dimensional data it makes sense to filter rows of a matrix by some condition defined on one of the columns whereas in vector space one works with reference vectors Many techniques exist for visualising multi dimensional data One of them is parallel coordinates They are the major topic of this thesis and are discussed in detail in Chapter 3 Parallel coordinates do not have any theoretical limit in the number of dimensions they can display at the same time they are only limited by very practical boundaries like screen space Other visualisation techniques discussed in the following subsections are also conceptually more limited with regards to the possibility of displaying an arbitrary number of dimensions all at once Note that the usefulness of visualisation techniques can be increased by viewing the same data set with different but synchronised methods on the same screen combining their distinct advantages Bertini et al 2005 Shannon et al 2008 This consideration has become an important aspect of information vi sualisation and in 2003 a dedicated annual conference has been launched the International Conference on Coordinated amp Multiple Views in Exploratory Visualization CMV Roberts 2007 Nevertheless as far as usability is concerned multiple views have also certain costs associated with them because there is less screen space available for individual visualisations and because th
58. Basic queries can be combined into more complex ones by logical AND OR and NOT operators Parallax also features automatic visual classification The most distinguishing feature of parvis is progressive rendering To enhance usability of the ap plication even for large data sets rendering is performed in two steps While the user modifies the display such as by dragging an axis or by selecting records with the mouse a fast and low quality no anti aliasing version of the visualisation is displayed immediately It is then replaced progressively by anti aliased lines which are generated in the background This way the system remains responsive while the user still performs operations with the mouse Work can be continued without interruption The other performance advantage gained from this is that most of the time the expensive generation of the high quality display can be terminated immediately as long as the user keeps moving the mouse pointer A further distinguishing feature of parvis are axis histograms which visualise the density of records on an axis interval parvis also allows the free choice of all colours in the parallel coordinates visualisa tion On the negative side there is no synchronised table view of the data set 3 5 4 Parallax Parallax is a parallel coordinates visualisation desktop application developed by Inselberg 2010a It is a commercial closed source software package Inselberg 2009 pp 382 416 gives a wal
59. Multiple Document Interface http msdn microsoft com en us library ms644908 Cited on pages 57 and 76 Microsoft 2010b Microsoft Visual Studio Express Build Cutting Edge Windows Applications http www microsoft com express Windows Cited on pages 54 and 110 Microsoft 2010c Registry Functions Windows http msdn microsoft com en us library ms724875 28v vs 85 29 aspx Cited on page 87 Microsoft 2011 Select multiple files or folders http windows microsoft com en US windows vista Select multiple files or folders Cited on page 29 Motif 2000 Motif http www opengroup org motif Cited on page 45 Mozilla Developer Network 2011 About JavaScript https developer mozilla org en JavaScript About_JavaScript Cited on page 47 Nash David 2006 The HYG Database http www astronexus com node 34 Cited on pages 16 and 32 Nimoy Leonard Lawrence Konner Mark Rosenthal Nicholas Meyer and Denny Martin Flinn 1991 Star Trek VI The Undiscovered Country 1991 Memorable quotes http www imdb com title tt0102975 quotes Cited on page 91 Novotny Matej 2004 Visually Effective Information Visualization of Large Data In Proc 8 Central European Seminar on Computer Graphics CESCG 2004 pages 41 48 http www cescg org CESCG 2004 papers 21_NovotnyMatej pdf Cited on page 17 Novotny Matej and Helwig Hauser 2006 Ourlier Preserving Focus Context Visualization in Par
60. Ordering Spacing and Filtering for Exploration of High Dimensional Datasets In Proc 9 IEEE Symposium on Information Visualization InfoVis 2003 pages 105 112 ISBN 0769519881 doi 10 1109 INFVIS 2003 1249015 http davis wpi edu xmdv docs infovis03_osf pdf Cited on page 19 Johansson Jimmy Matthew Cooper and Mikael Jern 2005a 3 Dimensional Display for Clustered Multi relational Parallel Coordinates In Proc 9 IEEE International Conference on Information Visualisation IV 2005 pages 188 193 IEEE Computer Society ISBN 078039464x doi 10 1109 IV 2005 1 http webstaff itn liu se jimjo papers IVO5 paperIV05 paf Cited on pages 20 and 21 Johansson Jimmy Patric Ljung Mikael Jern and Matthew Cooper 2005b Revealing Structure within Clustered Parallel Coordinates Displays In Proc 11 IEEE Symposium on Information Visualization InfoVis 2005 pages 125 132 IEEE Computer Society ISBN 078039464x doi 10 1109 INFVIS 2005 1532138 http webstaff itn liu se jimjo papers Infovis2005 johansso paf Cited on page 19 Johansson Jimmy Robert Treloar and Mikael Jern 2004 Integration of Unsupervised Clustering Interaction and Parallel Coordinates for the Exploration of Large Multivariate Data In Proc 8 IEEE International Conference on Information Visualisation IV 2004 pages 52 57 IEEE Computer Society ISBN 0769521770 doi 10 1109 1V 2004 80 Cited on page 17 Josuttis Nicolai 1999 The C Standard Library
61. Parallel Coordinates Exploratory Data Analysis with Parallel Coordinates and the Multi Dimensional Explorer Master s Thesis at Graz University of Technology submitted by Christian Hackl Institute for Information Systems and Computer Media IICM Graz University of Technology A 8010 Graz Austria 15 March 2011 Copyright 2011 by Christian Hackl Advisor Ao Univ Prof Dr Keith Andrews TU Parallelkoordinaten Empirische Datenanalyse mit Parallelkoordinaten und dem Multi Dimensional Explorer Masterarbeit an der Technischen Universitat Graz vorgelegt von Christian Hackl Institut fiir Informationssysteme und Computer Medien IICM Technische Universitat Graz A 8010 Graz 15 Marz 2011 Copyright 2011 Christian Hackl Diese Arbeit ist in englischer Sprache verfasst Begutachter Ao Univ Prof Dr Keith Andrews TU Abstract Parallel coordinates are an analysis technique in which multi dimensional data is visualised by ar ranging axes in parallel to each other on a plane Each dimension is represented by one axis Each data record is represented by a polyline connecting to one point on every axis The corresponding research field of information visualisation is first presented and numerous tech niques are discussed Parallel coordinates are then discussed in detail including issues of scalability and performance Previously developed software tools featuring parallel coordinates
62. S AS AAA 5 ee ee A E e cag in 7 Be Te at 4 pg 2a ze a _ 7 LE CALAS z 3 a Zn E ee RIA A e a 3 SN ee oo ea Z er an 1 ie 7 7 Teh Sate ee en Ma ae e a ur A 7 Z i 2 2 o gt u 1 SS oa ae DAA ee 2 ee er re cs ol A er 1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 al Year of Production gt ABCDFGHLMNPRS T WZ Actor ALL u AB C DFGHJKLM PRS TWZ Actress ALL Au ABC DFGHKLM PRS TWZ Director ALL au AB C DFGHJKLM PRS TWZ 0 Length 450 FE 0 450 Ratings WG E PG W PG 13 W R Films Shown 1455 Ach Copyright C 1993 HCIL Figure 2 6 The FilmFinder is a software application developed in the nineties of the 20 century using the scatter plot technique to visualise a data set about films Ahlberg and Shneiderman 1994 The horizontal and vertical axes correspond to the films popularity and age while the colour of the plotted points conveys information about film categories For example the red point in the bottom left corner shown here denotes a rather unpopular drama film from around 1925 Image copyright of the HCIL 2006 2 3 Multi Dimensional Data Visualisation 115 8 12 0 Gross purchasing power Figure 2 7 Vertical histogram feature of Multi Dimensional Explorer MDE the tool developed for this thesis applied to the data
63. Subsequently the virtual member function part iallyCreateVisualisation is called for each record and the likewise virtual member function visualisationLastStep is called to complete the visualisation VisualisationImplementationOpenGL can now cache the complete scene in pixels e If only focussing changed visualisation just calls its renderVisualisationToScreen func tion It first instructs VisualisationImplementationOpenGL to paste its pixels cache to the screen and calls the virtual member function createOverlay That function does nothing in the base class but is overridden in InteractiveVisualisation to handle features like focussing records under the mouse If for example a parallel coordinates polyline is focussed then the high lighted polyline is drawn on top of the entire cached scene The results of applying the overlay are not cached because overlay graphics are inherently temporary so there would be no sense in storing them persistently 5 2 Rendering TrueType Fonts in OpenGL OpenGL has no direct support for rendering text because text exists a higher abstraction level than the problems OpenGL solves and which are much closer to the graphics hardware Software engineers requiring text to be drawn with OpenGL therefore choose from a number of very different available solutions to meet their specific needs Khronos Group 2011b The solution employed by MDE is to use the FTGL library Maddock 2008 It enables progra
64. The green arrows on top can be clicked to flip the axes 4 2 7 Test Module Test contains MDE s unit tests Conceptually all other modules are tested it therefore depends on all of them Furthermore Test depends on UnitTest Llopis and Nicholson 2011 the unit test library used by MDE Every unit test s implementation resides in a cpp file of its own Header files in this module serve to reduce code duplication in cases where different unit tests need the same basic setup The vast majority of unit tests are for the modules Data and Data Input They are listed in Tables 4 9 and 4 10 respectively Other unit tests are listed in Table 4 11 4 3 Comparison With Other Tools MDE differs from other parallel coordinates software both from the user and from the developer point of view There are also many similarities though This is due to MDE combining features found in separate applications which in turn allows users to analyse data with a single application rather than having to switch between tools while working on the same data set Combining those features is not only an aggregation of functionality it leads to new ways of working with data because the set of features in a software application is more than the sum of its parts In summary what MDE contributes to the field of parallel coordinates is an integration of best user interaction practises found in previously developed tools Figure 4 10 shows the MDE application window and
65. Visualisation override private virtual methods and are implemented in terms of the pro tected non virtual methods of Visualisation Furthermore visualisation automatically creates and 68 4 Multi Dimensional Explorer A Bj b ScatterPlotVisualisation En oil a ete En sas reee c RecordTableData View d DimensionTableData View Figure 4 6 The concrete subclasses of MDE s abstract base class Dataview a paral lel coordinates drawn by ParallelCoordinatesVisualisation b a scatter plot drawn by ScatterPlotVisualisation Those two classes are indirectly derived from Dataview via Visualisation and InteractiveVisualisation On the other hand RecordTableDataView shown in c and DimensionTableDataView shown in d are direct implementations of textual representations of the data set maintains an instance of structure VisualisationStructure a structured collection of geometrical definitions explained in Subsection 4 2 6 MDE can likewise be configured to use different implementations for the visualisations it renders to the screen by subclassing the abstract visualisationImplementation base class and providing a fac tory to instantiate it by subclassing VisualisationImplementationFactory this is in accordance with the Factory Method design pattern described by Gamma et al 1994 visualisationImplementation itself is an extension of wxwindow The mod
66. a and Munich correspond to the light blue dots and to the light blue active but not selected polylines Additionally a thematic similarity map of the data is shown in the right top corner 35 36 3 Parallel Coordinates InfoScope is aimed at data with a geographical dimension such as data for a set of cities because the geographical location of records is shown on a map in an extra window The nature of the map and whether it is shown at all depends on the data Examples of demo data sets provided by Macrofocus include Prices and Earnings around the Globe UBS 2009 which uses a map of the whole world and Swiss Cities Ranking l Agefi 2003 which uses a map just of Switzerland The other additional visualisation provided is a thematic similarity map which is calculated so as to place cities with similar characteristics near to one another using techniques such as force directed placement Fruchterman and Reingold 1991 The central parallel coordinates visualisation component fixes the order of dimensions Axes cannot be reordered nor can single axes be removed from the view Nevertheless dimensions are grouped into categories this categorisation is reflected in the visualisation since the user can visually broaden a category The dimension axes belonging to the category are then separated by wider spacing while the other categories are narrower InfoScope distinguishes between active and selected records Arrow
67. a input It also provides an output facility selected records can be saved as a subset into a new okc file which can then be reopened later for further processing Dimension axes can be freely reordered and removed from the display but cannot be repeated Aside from manual reordering the tool also features a number of algorithms to order axes computationally Furthermore the space between axes can be stretched with the distortion feature Selecting records in the tool s parallel coordinates component is performed via combined interval definitions on each axis When the user moves the mouse over the visualisation the application s status bar displays the dimension closest to the mouse pointer along with the value corresponding to the mouse pointer s current vertical position Clicking then sets the upper or lower bound of the interval to be defined on that axis The intervals are represented by a grey polygon whose vertices lie on the axes Consequently the interior of the polygon contains the polylines of all selected records Figure 3 26 shows an example of this record selection method Up to four different selection polygons can be created 3 5 Software Applications 45 File Run Technique Options CombinedDist _ConbinedDist 1 _ attribute 0 attribute 1 131 no of results 7977 1620 no objects 16000 max of query 749 900 511 770 ARA min of query 262 483 346 541
68. acting with Computers 18 6 pages 1278 1309 doi 10 1016 j intcom 2006 03 006 Cited on page 26 Small Hugh 1998 Florence Nightingale s Statistical Diagrams In Florence Nightingale Museum Research Conference 1998 http www florence nightingale avenging angel co uk GraphicsPaper Graphics htm Cited on page 4 Sutter Herb 2001 Sutter s Mill Virtuality C C Users Journal 19 9 pages 53 58 ISSN 1075 2838 http www gotw ca publications mil118 htm Cited on page 54 Sutter Herb 2008 No Time for JAVA http www youtube com watch v xF0 LGoXtaw Cited on page 53 Swartzwelder John David Samuel Cohen Steve Tompkins and Bob Anderson 1995 3F04 Tree house of Horror VI http www snpp com episodes 3F04 html Cited on page 13 Swayne Deborah F Dianne Cook Andreas Buja Duncan Temple Lang Hadley Wickham and Michael Lawrence 2006 GGobi Manual http www ggobi org docs manual pdf Cited on page 43 Swayne Deborah F Dianne Cook Duncan Temple Lang Andreas Buja Nicholas Lewin Koh Heike Hofmann Michael Lawrence and Hadley Wickham 2007 The History of GGobi http www ggobi org history html Cited on page 42 Tel Tk 2010 Tcl Developer Xchange http www tcl tk Cited on page 44 Theisel Holger 2000 Higher Order Parallel Coordinates In Proc 5 h International Fall Work shop for Vision Modeling and Visualization VMV2000 pages 415 420 IOS Press ISBN 15860310
69. adapting its visualisations For example in MDE dimensions can be moved and flipped with simple mouse operations instead of requiring the user to open separate dialogue windows and menus Parallax lack of anti aliasing support means that there is no way to change the opacity of polylines whereas polyline opacity can be freely modified in MDE Situvis by Clear et al 2009 does not feature much user interaction and it is generally more diffi cult for the user to maintain an overview of the relationship between the parallel coordinates view and the details of the data it visualises The graph visualisation featured in previous versions of Situvis is absent from MDE However MDE uses the same approach to define filters on enumera tion dimensions by having the user click dots on an axis corresponding to individual elements of the enumeration In addition to that MDE also provides range filters for enumerated dimensions Picviz Tricaud et al 2011 is meant to be used in the context of network security while MDE is a general purpose visualisation tool for multi dimensional data Picviz generally offers few user interaction features compared to MDE It is not possible to set filters on dimensions or to move or flip axes The rendering backend does not use anti aliasing However the mechanism of selecting records by drawing a rectangle with the mouse is the same as in MDE GGobi Cook and Swayne 2007 offers statistics tools absent from MDE Its
70. ade known to the rest of MDE by allocating an instance of it in the constructor of DataViewFactoryGroup and adding it to its internal collection of factories MDE will then automatically put appropriate items on its main window tool and menu bars 116 B MDE Developer Guide Multi Dimensional Explorer a Ol xl File Views Window Help Figure B 5 With partiallyCreateVisualisation doing nothing TestVisualisation appears to the user as an empty window Visualisation function must be implemented for real partiallyCreateVisualisation is called by the visualisation base class every time a record has to be rendered The derived class TestVisualisation in this example uses the protected methods of Visualisation to implement the function such as drawLine Of drawRectangle The input data to be visualised is accessed and can be modified via dataHolder The goal of this guide is to create a visualisation displaying a horizontal line for every record This requires four Of Visualisation s protected functions e dataHolder to access the data e drawLine to draw a line setStrokeColour to set a line s colour e getOption to choose appropriate colours for different records Listing B 6 shows the source code for TestVisualisation to draw the corresponding lines This completes the developer guide Other features of MDE graphics drawing are found in Christian Hackl s thesis and in the source code docu
71. alisation Inselberg 2009 explains exactly 13 14 3 Parallel Coordinates a Orthogonal axes b Parallel axes Figure 3 1 In a point p lies at coordinates a and b corresponding to axes A and B The axes are orthogonal to each other In b p is still at coordinates a and b on axes A and B but the axes are now parallel to each other Consequently p is no longer represented by a point but by the line drawn from a to b how parallel coordinates can be used to extract information from data sets He also discusses some of the difficulties in applying parallel coordinates to very large data sets where the visualisation technique may not scale very well Those problems are discussed in detail in Section 3 2 of this thesis As far as usability issues are concerned Siirtola et al 2009 performed eye tracking studies on par allel coordinates visualisations finding that even inexperienced users quickly learn to use the technique effectively despite initial scepticism or uneasiness with the visualisation Following the teachings of d Ocagne the theory of parallel coordinates begins with the basic idea of duality between lines and points Any two dimensional point defined by orthogonal axes can also be visualised as a two dimensional line Expanding on this concept any n dimensional point can be visualised as a polyline consisting of n elements Imagine a two dimensional point p on a plane It has two coordinates one f
72. allel Coordinates IEEE Transactions on Visualization and Computer Graphics 12 5 pages 893 900 ISSN 1077 2626 doi 10 1109 TVCG 2006 170 Cited on page 17 126 Bibliography Oracle 2011 Java http www java com en Cited on pages 31 46 and 48 Organisation for Economic Co operation and Development 2011 OECD eXplorer interactive maps for regional statistics http stats oecd org OECDregionalstatistics Cited on pages 46 47 and 81 Pearson Karl 1895 Contributions to the Mathematical Theory of Evolution II Skew Variation in Homogeneous Material Philosophical Transactions of the Royal Society of London A 186 pages 343 414 doi 10 1098 rsta 1895 0010 Cited on page 8 Picviz 2010 The Picviz Language http trac wallinfire net picviz wiki PcvLanguage Cited on page 40 Piringer Harald Christian Tominski Philipp Muigg and Wolfgang Berger 2009 A Multi Threading Architecture to Support Interactive Visual Exploration IEEE Transactions on Visualization and Com puter Graphics 15 6 pages 1113 1120 ISSN 1077 2626 doi 10 1109 TVCG 2009 110 Cited on page 6 Playfair William 2005 Playfair s Commercial and Political Atlas and Statistical Breviary Cambridge University Press ISBN 0521855543 doi 10 2277 0521855543 Cited on page 3 Popper Karl 1996 Alles Leben ist Probleml sen Piper ISBN 3492037267 Cited on page 3 Ramani Nandini Andrew Shellshear Craig Northway Vince
73. amin Qu Weiwei Cui and Baoquan Chen 2008 Visual Clustering in Parallel Coordinates Computer Graphics Forum 27 3 pages 1047 1054 ISSN 0167 7055 doi 10 1111 j 1467 8659 2008 01241 x Cited on page 17
74. and weaknesses making it an ideal starting point for anyone willing to try exploratory data analysis with parallel coordinates and looking for the proper tool to solve the task at hand At the same time there is a great variety of applications of parallel coordinates This is demonstrated by this thesis as well because the software reviewed covers many different domain areas ranging from 91 92 6 Conclusion and Future Work conflict analysis network security and regional statistics to medical science Other than that technical aspects of the implementations differ too Some of the tools are classical desktop applications others are embedded into web pages for online access In short this thesis shows parallel coordinates versa tility both from the end user s and the software engineer s point of view Performance turned out to be a common problem in the parallel coordinates software reviewed However as previously discussed user interaction is becoming ever more important in information visualisation The best clustering technique has yet to be found As a result existing software tools do not handle large data sets in a satisfactory manner for the end user of an application The Multi Dimensional Explorer MDE software was developed for this thesis and was discussed in Chapter 4 MDE is faster than traditional parallel coordinates software by its choice of implementation tools C and OpenGL and by applying a variety of
75. andard Renderer does not require OpenGL but will perform much slower OpenGL with anti aliasing disabled will usually not bring performance gains justifying the requisite loss of alpha transparency The Reset All Options button is used to reset all options in MDE to the factory defaults A 6 Options 107 J T Bad T Colc W Cole co F Cole TT Rer T Fon For T For I Font E t I Font I Colo I Cole T Cole I Toc Figure A 16 General options can be overridden in MDE s options dialogue window This exam ple shows the user overriding the colour of selected records specifically for parallel coordinates views By default values inherited from the General page are used in order to override them the check box to the left has to be clicked first Colours can be specified either by typing their hex adecimal RGB code or by clicking the coloured button on the right which brings up a standard colour dialogue window 108 A MDE User Guide Appendix B MDE Developer Guide What you write and how you write it can change your life Beppe Severgnini Litaliano Lezioni semiserie 2008 p 10 This guide is meant for future developers of Multi Dimensional Explorer MDE a software appli cation developed in 2011 by Christian Hackl for his Master s thesis Exploratory Data Analysis with Parallel Coordinates and the Multi Dimensional Explorer at Graz University of Technology It ex plains how
76. andling Large Data Sets Figure 3 4 shows what happens when the data set to be visualised by parallel coordinates is very large the display becomes cluttered with polylines up to a point where the entire visualisation becomes unusable because lines overlap and no more patterns can be seen Artero et al 2004 Wegman and Luo 1996 The problem can be mitigated somewhat by reducing the polylines opacity as shown in Figure 3 5 When each polyline is rendered with reduced transparency patterns previously hidden to the eye can once again be recognised because areas of many overlapping lines will appear darker than areas with fewer lines Holten and van Wijk 2010 Reducing polyline opacity is a simple way to mitigate the problem of cluttered parallel coordinates displays but may not always be sufficient Fua et al 1999 argue that the root of all such problems is the idea of mapping every record of the data set to one particular graphical element and that visualisation techniques based on such direct mapping will always scale badly in face of large input data Novotny 2004 refers also to performance problems with software applications having to render overly complex graphics Therefore clustering may be needed for many large data sets Novotny and Hauser 2006 name a few thousand as the number of records visually comprehensible with parallel coordinates whereas hundreds of thousands of records are an indication that problems will a
77. applet Interaction in this tool is restricted to moving axes vertically zooming axes and designating one axis as the reference for colouring of records 49 50 3 Parallel Coordinates Google docs cereal Private only to me Updated 8 days ago by christian hackl Saved Share y File View Help Edit Settings Publish Add Gadget to iGoogle more v Cylinders Horsepower Weight Sheet Gadget GE Figure 3 32 The parallel coordinates component of Liquid Diagrams by Andrews and Lessacher 2010 a set of gadgets to be used in Google Docs Given the nature of the tool it can be used anywhere where web access exists It features extensive support for user interaction and SVG export This figure shows the tool visualising the well known cars data set Ramos and Donoh 1983 3 5 Software Applications 51 There are also some options for customisation of the visualisation such as colours and fonts for various labels The most powerful of those options is to choose the colour scheme of the polylines such that every record is given an individual colour This works well for small data sets with few records because the colours are more easily distinguishable Performance is an issue with Liquid Diagrams Since the gadget is implemented in Flash and runs inside a web browser it cannot always deliver the speed and responsiveness required for fluent interactive analysis of data On the other hand there is also the unique fe
78. are reviewed and their strengths and weaknesses analysed The interactive features necessary to perform exploratory data analysis with parallel coordinates are illustrated A new general purpose information visualisation tool the Multi Dimensional Explorer MDE was developed in C and OpenGL It facilitates exploratory data analysis with high performance and a variety of user interaction features The thesis documents the modular software architecture of MDE User and developer guides for MDE are included as appendices Kurzfassung Parallelkoordinaten sind ein Analyseverfahren bei dem mehrdimensionale Daten visualisiert wer den indem Achsen parallel zueinander auf einer Ebene angeordnet werden Jede Achse steht fiir eine Dimension und jeder Datensatz wird mit einer Polylinie dargestellt die durch je einen Punkt auf jeder Achse l uft Erst werden der dazugeh rige Forschungsbereich der Informationsvisualisierung vorgestellt und ver schiedene andere Techniken besprochen dann werden Parallelkoordinaten im Detail er rtert samt Be trachtungen zu Skalierbarkeit und Performance Bisher entwickelte Parallelkoordinaten Software wird besprochen und deren St rken und Schw chen werden untersucht Ebenfalls erl utert werden die f r die empirische Datenanalyse mit Parallelkoordinaten n tigen Interaktionsm glichkeiten Eine neue universell verwendbare Informationsvisualisierungs Software namens Multi Dimensional Explorer MDE wurde mit C
79. ational Conference SUGI 16 http www math yorku ca SCS sugi sugil6 paper html Cited on page 8 Friendly Michael and Daniel D Denis 2001 Milestones in the History of Thematic Cartography Sta tistical Graphics and Data Visualization http datavis ca milestones Cited on page 4 Fruchterman Thomas M J and Edward M Reingold 1991 Graph Drawing by Force directed Placement Software Practice and Experience 21 11 pages 1129 1164 doi 10 1002 spe 4380211102 http www cs ubc ca local reading proceedings spe91 95 spe vol21 issuell spe060t f pdf Cited on page 36 Fry Ben and Casey Reas 2010 Processing org http processing org Cited on page 40 Fua Ying Huey Matthew O Ward and Elke A Rundensteiner 1999 Hierarchical Parallel Coordi nates for Exploration of Large Datasets In Proc 10 International IEEE Visualization Conference Vis 99 pages 43 50 doi 10 1109 VISUAL 1999 809866 Cited on pages 17 and 18 Gamma Erich Richard Helm Ralph Johnson and John Vlissides 1994 Design Patterns Elements of Reusable Object Oriented Software Addison Wesley Professional ISBN 0201633612 Cited on pages 53 57 59 63 65 68 and 70 Goel Amit 1999 VizCraft A Multi Dimensional Visualization Tool for Aircraft Design Master s The sis Department of Computer Science at Virginia Polytechnic Institute and State University Virginia Tech http www amitgoel com projects vizcraft Cited on page
80. ations For example the scatter plot has no notion of polyline opacity so there is also no setting for it Figure A 12 shows an example of record states being synchronised across different views The two dimensions to be visualised in the scatter plot are chosen with the controls at the bottom of the window This is also the location of the SVG Export button the zoom control and the Options button to fine tune some display options A 4 Records View The records view is a textual tabular representation of the input data It therefore does not feature zoom ing or SVG export However record states and its colour scheme are synchronised with other visuali sations Figure A 13 shows a Records window displaying unused records in grey selected ones in blue and the focussed one in red just like in the parallel coordinates and scatter plot visualisations Furthermore the order of columns in the records table is synchronised with that in the parallel coor dinates display The table can be sorted by clicking on one of its headers Note that the order of records is an attribute specific to the records view Other views have no notion of an order relationship between records The records view also features incremental search Users can type text in the text field at the top right of the window As they type the records shown in the table are restricted to those with at least one field containing the typed text Another useful feature of the records v
81. ature of exporting the visualisation as an SVG file This makes it particularly easy to embed a parallel coordinates visualisation inside other documents or presentations or to edit it outside of Google Docs 52 3 Parallel Coordinates Chapter 4 Multi Dimensional Explorer We are the champions O Bjarne And we ll keep compiling till the end Herb Sutter at SD Best Practices 2008 This chapter is about the software application Multi Dimensional Explorer MDE which was writ ten for this thesis It implements parallel coordinates a visualisation technique discussed in Chapter 3 that chapter also discusses previously developed tools MDE learned from them by building upon their strengths and useful ideas while at the same time avoiding some of their pitfalls Although MDE is a general purpose visualisation tool designed to be extensible with additional visualisation schemes and not dependent on any concrete visualisation scheme its focus is on parallel coordinates It is a desktop application with a graphical user interface supporting interaction that is not only displaying static images MDE is designed to be as cross platform as possible Performance is a major requirement which is why OpenGL Shreiner 2009 Khronos Group 2011a was chosen as the main rendering engine Section 4 1 outlines the software design principles of MDE and documents the environment in which it was developed Section 4 2 discusses the individ
82. ay of MD VS a tool developed by Mader 2007 visualising a data set about prices and earnings in different cities around the globe provided by UBS 2009 The data set was turned into a comma separated value file suitable for opening in MD VS with the InfoScope tool made by Brodbeck and Girardin 2003 Data points clustered in the centre of the left island displayed as red circles are Bratislava Copenhagen Prague Vilnius and Warsaw indicating that these cities are similar in terms of prices and earnings Chapter 3 Parallel Coordinates Take that you lousy dimension Chief Wiggum in The Simpsons Episode 3F04 1995 Parallel coordinates are a way to visualise multi dimensional data by displaying axes in parallel rather than orthogonal to each other Parallel coordinates belong to the computer science field of multi dimensional data visualisation which was introduced and discussed in Chapter 2 Section 3 1 explains the theory of parallel coordinates and their roots in mathematics Section 3 2 discusses problems of parallel coordinates handling large data sets and possible solutions thereof Sec tion 3 3 introduces some variations on the general idea of parallel coordinates proposed by the scientific community Section 3 4 explains the operations provided by interactive parallel coordinates visualisation for data analysis in high dimensional space Finally Section 3 5 gives an overview of existing software solu
83. bar or by selecting the appropriate item from the File menu There is also a tool bar button and a menu item for generating random data although this feature might be more useful for developers testing visualisations rather than actual end users Upon selecting a CSV file to be opened MDE opens a dialogue window like the one shown in Figure A 1 in which CSV parsing parameters can be overruled in case the application did not succeed in automatically detecting the correct separators comment characters quoting characters decimal point characters and escape characters Upon clicking OK the data is loaded and will automatically be visualised as parallel coordinates Every view of the data resides in its own MDE subwindow A new view can be opened by clicking one of the Parallel Coordinates Scatter Plot Records and Dimensions buttons on the tool bar or by selecting the corresponding items in the Views menu Almost all of MDE s user interaction takes place in the view subwindows They are discussed in subsequent sections 95 96 A MDE User Guide Import CSV File Figure A 1 The Import Wizard which MDE shows to the user upon selection of a CSV file to be opened for visualisation If MDE did not succeed in detecting the correct separators com ment characters quoting characters decimal point characters or escape characters the user can override these parameters The left part of the window shows the plain te
84. bility order minimum and maximum values and filters All these attributes of a dimension can be changed in this window and the effect of any change becomes immediately visible in all other win dows Figure A 15 shows the modification of a dimension s minimum and maximum values causing the corresponding axis in the parallel coordinates visualisation to be updated The bottom part of the window consists of controls with which the user can update remove or add filters Although the parallel coordinates visualisation features the same mechanisms the dimensions window is more appropriate for setting precise filters For example setting a filter s upper limit to 13 14 is difficult in the parallel coordinates window whereas in the dimensions window the exact value can be typed in Unlike other views the dimensions window has no special options associated with it A 6 Options All visual attributes of MDE can be customised in options dialogues The options dialogue window is opened via the Options item in the Views menu It presents a hierarchy of option pages with general options at the top the parallel coordinates scatter plot and records views have their own pages the dimensions view does not because there are no visual options associated with it A 6 Options 105 Dimensions 10 x LAW Tima ram iend Fae 1 len nf rien mM vacation paid working days per year Type Integer Real Enumeration Min 0 0 Max 100
85. ceeded Figure B 2 The Visual C development environment for MDE after having opened the main solution file and performed a build with the Release configuration This guide assumes Microsoft Visual C Express Edition as the development environment No full version of Microsoft Visual C is needed Express Editions are provided free of charge by Microsoft and contain no technical restrictions in the compiler and debugger Figure B 2 shows mde s1n opened in Visual C Express Edition 2010 and built with the Release configuration B 2 Preparing a New Visualisation The new visualisation shall be named Testvisualisation residing in namespace presentation and it will be a subclass of Visualisation Before implementing it however header and source files for it must be added to the solution To do so a developer first has to undertake these steps 1 Create a header file called test_visualisation h for the new visualisation in the presentation subdirectory The easiest way to achieve this is to use the Solution Explorer as shown in Fig ure B 3 by right clicking on Header Files item of mde presentation and selecting Add New Item In the dialogue window which appears Header File h must be chosen 2 Create the associated source file test_visualisation cpp repeating the step from above but choosing instead the Source Files item of mde presentation and choosing the C File cpp entry in the
86. chniques to solve problems caused by overlapping polylines 3 5 Software Applications This section gives an overview of existing software solutions for parallel coordinates visualisations Each application is introduced separately and particular features strengths and weaknesses are highlighted Parallel coordinates are simple enough to implement but still each software application reviewed in this section turns out to be different e Features Applications may offer certain features not found in competitors For instance Parallax features automatic visual classification XmdvTool can animate data along a time dimension and GGobi has extensive statistical tools e Purpose Some of the applications also differ in their general scope and purpose InfoScope is particularly aimed at exploring data with a geographical aspect whereas Picviz was developed for use in the field of network security Situvis was developed in the context of conflict analysis e Interaction Finally applications differ in the way users can interact with the display For exam ple parvis features two phase rendering to keep the software responsive while the user manipulates the visualisation Caleydo has angular brushing Table 3 2 lists the reviewed software applications Software which turned out to be more feature rich with regards to parallel coordinates was given more space than the other tools When software is available for download or direct online use on the
87. cords 49 used 8 selected 52 dmensions Figure A 13 MDE s records view is a table in which records are presented in textual form The states of records are synchronised with all other views This example shows selected records in blue unused ones in grey and the focussed one Barcelona in red This matches exactly the colours in the parallel coordinates window above 104 A MDE User Guide Parallel Coordinates Amsterdam 29 Copenhagen 115 8 73 Zurich 106 8 110 5 10 Figure A 14 MDE s records view allows users to display only relevant records in the table For example if users are interested only in seeing entries of used records they can click on Show only used This image shows the user setting a ranged filter on the Gross purchasing power dimension in the parallel coordinates window leading to only Copenhagen and Zurich being included The records view now only shows these two records dimensions using the handles in the parallel coordinates window This gives an immediate overview of which records are included in the range and which are not Figure A 14 shows what this looks like in MDE Visual properties of the records view can be customised by clicking on the Options button A 5 Dimensions View MDE s dimensions view is used to fine tune information about dimensions rather than individual records It is synchronised with the other views in terms of data pertaining to entire dimensions name type vis i
88. cords windows displaying green table rows for selected records 2 Open the options dialogue select the Parallel Coordinates entry and override the Colour Of Selected Records option only for parallel coordinates windows Figure A 16 shows how to do this By default all inherited general options use the original value in order to override an option and specify a custom colour the check box to the left has to be clicked first This will lead to all parallel coordinates windows displaying green polylines for selected records but not cause any change in the other windows 3 Click the Options button in the current parallel coordinates window and override the option there This will lead to only the current window being affected Other parallel coordinates windows and parallel coordinates windows opened later are not affected The modification is not stored in options xml either However a user can always choose to make the change persistent and apply 1t to all parallel coordinates windows by clicking the Set As Default button In MDE colours can be specified either by typing their hexadecimal RGB code or by clicking the coloured button on the right which brings up a standard colour dialogue window Fonts are specified by choosing a font name from a list For the parallel coordinates and scatter plot visualisation the rendering backend can be chosen too However the default OpenGL backend will usually suffice The alternative called St
89. cted by drawing a rectangle with the mouse as shown in Fig ure A 9 When the user moves the mouse pointer over a polyline it is focussed by drawing it in red After some delay 1 second by default tooltips showing the corresponding record s actual values on each axis appear as well as shown in Figure A 10 A 3 Scatter Plot View A scatter plot is a two dimensional visualisation technique for multi dimensional input data Any two of the input data dimensions are chosen by the user to represent the horizontal and vertical axes of the plot Data points are then plotted according to their values in the two selected dimensions Figure A 11 shows MDE s scatter plot component MDE synchronises the states of records across all of its views Records selected or turned unused in the parallel coordinates visualisation are selected or unused in the scatter plot view as well and vice versa The same applies to focussing records and to hiding entire dimensions However there 102 A MDE User Guide Net purchasing power Gross purchasing power w Y Axis Net purchasing power hd Options Figure A 11 A scatter plot is a standard 2D visualisation technique in which two dimensions of a multi dimensional data set are selected to represent the horizontal and vertical axes of the window Data points are then plotted on the window according to their values in the selected dimensions are also properties specific to individual visualis
90. d time required for 1 kg of rice and normal local rent medium 3 4 Using Parallel Coordinates a Polylines cross ing an axis at the same point b Curves crossing an axis at the same point Figure 3 9 Using curves instead of polylines for parallel coordinates a technique introduced by Graham and Kennedy 2003 a shows a problem with polylines in traditional parallel co ordinates visualisations it is difficult or outright impossible to tell which lines belong to which polylines because many polylines may cross axes at the same point For example does line 2 belong to polyline A or to polyline B b show curves solving the problem intuitively a user can immediately and visually observe that curve segment 2 belongs to polyline B The images are adaptions of example diagrams in Graham and Kennedy 2003 23 3 Parallel Coordinates Cities Gross purchasing power Net purchasing power Prices excl rent Amsterdam 95 3 85 5 83 0 Athens 61 8 63 3 72 7 Auckland 64 8 70 8 62 3 Bangkok 18 4 24 1 58 7 Barcelona 65 7 73 6 83 5 Beijing 21 7 24 1 57 4 Berlin 93 5 89 4 81 0 Bogota 32 3 37 9 47 1 Bratislava 35 7 39 5 59 4 Brussels 97 2 88 5 84 7 Budapest 34 2 29 6 54 4 Buenos Aires 30 5 34 9 50 4 Bucharest 33 5 33 7 46 0 Caracas 22 0 27 7 91 0 Chicago 97 9 96 3 82 0 Delhi 18 2 21 8 37 6 Doha 30 3 41 8 67 6 Dubai 4
91. d Development 2011 too is rooted in a particular domain namely statistics of regions around the world MDE is general purpose and thus accepts anything as input data while special measures have to be taken to load custom data into OECD eXplorer which will still try to interpret it as OECD statistics data The other major difference is that OECD eXplorer is implemented as an online Flash application whereas MDE is a native desktop application OECD eXplorer s mean line and histogram features have been inte grated into MDE However in OECD eXplorer groups of records cannot be selected by a drawing a rectangle but are selected by clicking on the axis histograms Protovis Bostock and Heer 2009 offers almost none of the interaction features of MDE From an implementation point of view its use of JavaScript and SVG is radically different from MDE which is a desktop application written in C and using OpenGL However MDE allows users to export visualisations as SVG files VizCraft by Goel 1999 offers almost none of MDE s features either It is targeted at aircraft design and implemented as a Java applet in contrast to MDE s general purpose approach and C implementation Liquid Diagrams Andrews and Lessacher 2010 is an online tool integrated into Google Docs which is yet another example of an approach radically different from MDE This difference in implementation also means that while Liquid Diagrams have the advantage of be
92. da eae aud 9 FilmPinder saosi a canoe see sae na a bee Gad ee ew Ewe ae 10 Histogram esi 4 aa 0 a la ea eb de Se ee we re ce 11 Slat Plots sacs oak u Gal we ara Se dw Got a Se oe Ge aw eo Ro eS 12 Vislslands coo A ar dete oe OE A hk we 12 2D Points on Orthogonal and Parallel Axes 2 2 ee ee 14 4D Points on Parallel Axes o ee 15 Parallel Coordinates in MDE 02 02 pe ee 15 Cluttered Parallel Coordinates o e 16 Cluttered Parallel Coordinates With Reduced Opacity 2 2 22 22 nn 16 Hierarchical Parallel Coordinates o o ee 18 Parallel Sets y 3 0 Be Rw u win ar ana 20 3D Parallel Coordinates 21 Curves for Parallel Coordinates 2 2 22 2 Co on n nn 23 Cities Data Set with Parallel Coordinates 2 22 on on nn 25 Removing Axes in Parallel Coordinates o o e e 25 Moving Axes in Parallel Coordinates 2 2 0 2 0 00000 000000 27 Flipping Axes in Parallel Coordinates 0 0 2 0000000004 28 Parallel Coordinates Record Details 0 nn 28 Angular Brushing s o e ay etaa i ea a d a a a a a A a a a a a aLe aa a 29 Colour Shading ios e aa 2 nase ea e Gah ee a e eo a how eg AS 30 Parallel Coordinates Zooming ee ee 31 Parallel Coordinates Focussing oaa ee 32 ADAYA S hae oe eA Ro DA dk a ee ae a ie doh bd 34 TINTOS COPE ceci oe ala Se as a a de ek a 35 PALVIS ur bos wl aoe eR ORE Bee ee ae ee ee ee dee
93. din 2003 too has a similar user interface Dimension filters are set by dragging triangles on axes however group selection of records is not as easy to achieve as in MDE with its simple selection rectangle approach However MDE lacks InfoScope s special support for geographical dimensions parvis Ledermann 2003 features axis histograms very much like those found in MDE Its multi phase rendering of the parallel coordinates visualisation however takes a different approach parvis draws a non anti aliased preview for visualisation areas about to be modified by current user actions and does not redraw areas which will not be touched MDE acknowledges that redrawing of same graphics on the same area is a performance bottleneck but defines multiple rendering phases which are cached internally as arrays of pixel data Anti aliasing can be globally turned off as an option though this has turned out to be of little use in terms of higher performance MDE s approach is more abstract and can thus be used to optimise other visualisation methods as well Parallax Inselberg 2010a offers more sophisticated record selection techniques than MDE such as angular brushing It has also a unique approach to the handling of empty fields in the input data which MDE simply transforms to 0 Setting filters on dimensions called interval queries in Parallax works in a similar manner MDE however uses more interaction oriented means of dy namically
94. ds may cause similar problems but can be approached in different ways Jing et al 2003 discuss ways of clustering dimensions to reduce visual overload They apply their technique not only to parallel coordinates but also to other visualisation methods 3 3 Variations The basic idea of parallel coordinates is to draw polylines corresponding to records in a data set on a 2D display Building on this basic idea other similar visualisation techniques have been devised The following subsections give an overview of variations on the original parallel coordinates theme 3 3 1 Parallel Sets Parallel sets Bendix et al 2005 adapt parallel coordinates for enumerated dimensions Parallel coor dinates usually deal with continuous numeric data types having an obvious order defined for them and no fixed number of allowed values In contrast enumerated dimensions such as Gender or Country usually have no natural ordering relationship is Austria less than Italy and define a fixed number of values a record in the data set may have Although enumerated dimensions can be normalised in order to be treated like dimensions of numeric type Mader 2007 parallel sets take a different approach altogether Just like parallel coordinates parallel sets display axes arranged parallel to each other However parallel sets do not plot points on the axes which are then connected by polylines instead rectangles of equal width and diffe
95. e ee ee Pe nee 89 A l MDE Opening a CSV File 0 00 0000000022 eee ee 96 A 2 Parallel Coordinates Concept 2 0 2 nn 98 A 3 MDE s Parallel Coordinates GUI Controls 2 2 2 2 2m nenn 98 A 4 Selecting Parallel Coordinates AxesinMDE o o 99 A 5 Interactive Elements on Parallel Coordinates Axis in MDE 99 A 6 Flipping a Parallel Coordinates Axis in MDE o o 100 A 7 Set Filters ima MDE 2 a ee ar anne re 100 A 8 MDE Axis Histograms ee 100 A 9 Selecting Records in MDE CL nommen 101 A 10 Focussing Records in MDE 2 2 2 oo none 101 Add MDE Scatter Plot os u 2 4 Sa ale De en een en A 102 A 12 Record Synchronisation in MDE 2 Coon nn 103 A 13 Records Table in MDE 2 2 nn 103 A 14 MDE s Records Table Showing Only Used Records 2 2220er 104 A 15 MDE s Dimensions View ee 105 A 16 MDE Options Dialogue ee 107 B 1 MDE Test Visualisation 2 2 a 110 B 2 MDE s Development Environment 0 0 0 00000 eee 111 B 3 Adding New MDE Visualisation Header File 004 112 B 4 New Visualisation AppearinginGUl o o e 114 B 5 New Empty Visualisation Window in MDE o 116 List of Tables 3 1 Cities DataSet 56 4 6G ORS 2 u ae u Be Ee oe aed ee 24 3 2 Parallel Coordinates Software Summary 000000 0b ee 33 4 1 MDE
96. e multitude of visualisa tion techniques results in increased complexity for the end user Baldonado et al 2000 give guidelines a software designer should take into account when deciding on whether or not to show different visuali sations of the same data set on the same screen 2 3 1 Scatter Plots A scatter plot is a two dimensional visualisation of multi dimensional information each axis of the display corresponding to one of the dimensions in the data set Data points are entered on the two dimensional area according to their values in both dimensions The points themselves can be made to convey information about additional dimensions as shown by Cleveland and McGill 1984 Examples of such additional information channels include e Point Size For example in a data set of cities larger points may indicate cities with greater population 8 2 Information Visualisation e Point Colour For example the blue tone of a point s colour may correspond to the city s total area e Point Form For example dots may indicate cities in Europe rectangles cities in America A different approach to overcome the two dimension limit of scatter plots is to create a set of each x y combination and display the resulting plots next to each other in a matrix This technique is called scatter plot matrix Elmqvist et al 2008 Figure 2 5 shows the visualisation of a data set about prices and earnings in different cities around the globe
97. e required to earn enough money for 1 kg of bread The selected record is shown in blue so that the user can distinguish between selected and unselected records The terminology is sometimes inconsistent in the literature Inselberg 2009 calls the operation querying and gives it a more general meaning He defines a query as any command the user can issue to parallel coordinates software to change the display such that desired information can be gained visually which may or may not involve a change in the appearance of individual records Note that the word is not to be understood in the classical computer science sense it has nothing to do with query languages such as Structured Query Language SQL or even with retrieving records from a database Instead the word denotes visual interaction mechanisms provided by the user interface Other sources call the operation brushing Ward 1994 For reasons of consistency the term selection will be used throughout this thesis Selection or brushing is preferably achieved by dragging the mouse to enclose desired line segments within a box The enclosed records are highlighted as selected For the aggregation of different selec tions it makes sense to use additional keyboard actions as well for example when the shift key is pressed while a selection rectangle is being drawn with the mouse the enclosed records might be added to the previous set of selected records rather than replacing
98. e software features various visualisation techniques in addition to parallel coordinates Figure 3 27 shows the application running on Linux and displaying three dimensional data The input format is a plain text file in which dimensions are separated by spaces and records by line breaks This is the format of the example data file which comes with the software no further documentation is given on file input formats 46 3 Parallel Coordinates lal x File Data View Window Help ES T rita El Ones ao 2 3 MO A ll oc Be Ba Ba te EE Ce Loge Eo Lo eke Ke Seek Loge Soe Es HAG SEARRAS El EE Serre EE eer ri IF Fi o 5484 i E 34d 388 ddr coca caco Figure 3 28 Caleydo Lex et al 2010 an OpenGL based Java tool aiding in the analysis of multi dimensional data in a medical domain gene expression data Its parallel coordinates component offers various interaction features including angular brushing The record currently under the mouse is focussed yellow in this image and tooltips with the record s values appear VisDB offers only limited user interaction for its parallel coordinates component Records cannot be selected or brushed Axes cannot be moved or hidden The visualisation s size is fixed The application does support reducing the number of records displayed and vertically stretching any of the axes 3 5 10 Caleydo Caleydo is a
99. e these problems a generic software architecture for implementation of information visualisation techniques is proposed In that architecture the main thread delegates each instance of a visualisation to a dedicated visualisation thread Wide spread use of multi threading in visualisation applications is hindered by the fact that concur rency itself is always very complex to handle in software engineering Thread safety is particularly hard to achieve when underlying software modules do not provide sufficient thread safety guarantees For ex ample the popular OpenGL graphics rendering library Shreiner 2009 Khronos Group 201 1a which is often used to implement high performance information visualisation software requires developers to take a number of precautions to ensure the robustness of software when multi threading is present Apple 2010 2 3 Multi Dimensional Data Visualisation Input data to be visualised comes in different forms and variations One of them is multi dimensional data as explained by Andrews 2010 Such data consists of n records and m dimensions Therefore the data can be seen as a matrix an item d of the matrix indexed by n and m as shown in Figure 2 3 A concrete example would be a matrix for cities each city record having the dimensions Name Area Elevation and Population as shown in Figure 2 4 Multi dimensional data is not just a synonym for vector space data The most important dif
100. e to stay Let us apply our knowledge in other areas of software engineering to make them faster better and more usable 94 6 Conclusion and Future Work Appendix A MDE User Guide Just TU it Motto of Graz University of Technology alias TU Graz This is the user manual of the Multi Dimensional Explorer MDE software an interactive informa tion visualisation desktop application for the exploratory analysis of multi dimensional data MDE was developed in 2011 by Christian Hackl for his Master s thesis Exploratory Data Analysis with Paral lel Coordinates and the Multi Dimensional Explorer at Graz University of Technology It documents MDE from the end user s point of view Future developers of MDE who might want to extend the tool should instead use the MDE Developer Guide No setup is needed for MDE All it takes to use the software are its exe file on Windows and a CSV file to be opened There is no installer for MDE A file in which program options are stored persistently options xml is created automatically in the same directory as the executable MDE does not modify the Windows registry All examples given in this manual use the well known cities data set by UBS a collection of informa tion about prices and earnings around the globe turned into a CSV file using the conversion mechanism of Macrofocus InfoScope A 1 Loading Data Data is loaded into MDE by pressing the Open File button on the tool
101. eadXml TestRecentFiles Tests application RecentFiles TestRecordIteration Tests presentation RecordIterationStrategy estTestOptionsTree Basic tests for options Tree Ite rator TestVisualisationElement Tests C standard algorithms on Algorithms presentation structure VisualisationStructure TestVisualisationElementLine Tests presentation structure Intersection intersectsRectangle TestVisualisationElementOrdering Tests order of presentation structure VisualisationElement elements TestVisualisationElementType Tests type equality of presentation structure Equality VisualisationElement TestVisualisationElementValues Tests return values of presentation structure VisualisationElement methods TestVisualisationStructure Tests iterators of presentation structure VisualisationStructure es tVisualisationStructureRecords Tests record handling Of presentation structure VisualisationStructure Table 4 11 The unit tests in MDE s Test module testing components of modules Application Presen tation Presentation Structure and Options 80 4 Multi Dimensional Explorer mouse filters are set on dimensions by vertically dragging triangles on an axis However MDVS uses standard operating system functions to draw its graphics and is thus much slower than MDE which uses an OpenGL backend InfoScope Brodbeck and Girar
102. ected and selected records Dragging Axes The horizontal position of each axis can be changed by dragging and dropping the header text with the mouse Tooltip Information When the user moves the mouse pointer over a record and leaves it there for a few seconds the record s actual data values are displayed on each axis 3 5 2 InfoScope InfoScope is a commercial closed source Java desktop application made by Brodbeck and Girardin 2003 and distributed by the Macrofocus company A free test version can be obtained from the com pany s web site The test version however can only be used with demo data sets provided by Macrofo cus Figure 3 20 shows the tool For data input the software uses proprietary mis files It is a binary format but the professional version of InfoScope provides InfoScope Distiller a tool for conversion of simple text files into mis files Macrofocus 2010 Data sets can be read by opening files from the local file system or by down loading them from Macrofocus servers directly from within the application 34 3 Parallel Coordinates EB ParallelCoords 1 Test Table 2 Ox Zoom 13 I 13 IV Antidliasing Vectornumbers VectorCluster Category Created Importance Quality opened vis a All records are used none are selected BB ParallelCoords 1 Test Table 3 ol x zoom 13 13 a Antialiasing Vectormumbers J VectorCluster
103. ed polygon encompassing the desired points Parallax features a comprehensive structured querying mechanism which can be viewed as a two layered system e On the first layer queries can be named and be associated with a particular colour Any number of such queries can be created The results of each query that is the colouring of records can be shown separately or in combination with those of other queries e On the second layer a query is itself composed of queries Thus the word query actually has a double meaning in Parallax Queries on this second layer comprise interval pinch angle and relative complement queries A pinch query is defined by an upper and lower arrow placed anywhere between two axes As the horizontal position of the arrows may be moved as well this must actually be considered an invisible selection line Those records whose polylines intersect with the invisible line are selected The interval query can likewise be considered a special case of the pinch query in which upper and lower bound arrows have to be placed exactly on an axis Angle queries defined between two axes constitute a range of angles An angle is measured from the left axis line For example the minimum allowed angle might be 30 degrees the maximum allowed angle 45 degrees Now if a polyline segment between the two axes has an angle of 35 degrees from the left line then it is included in that range and the record i
104. en the mouse pointer is hovering above any of its polyline segments Focussing means highlighting the polyline with another different colour not the one used for selected records and showing the corre sponding record s individual data values on the crossings between polylines and axes Figure 3 18 shows a focussed record in a crowded parallel coordinates visualisation 3 5 Software Applications 31 cities2009 csv Multi Dimensional Explorer Parallel Coordinates 18 x e Ele Views Window Help 181 x E _Paralel coordinates scatter Piot Records Dimensions El y N N u N Nx SN N N S N N X lt N NN N QO AY N b AQ AQAA N N Y QA 7 N gt XX SAMY x N SOSA QS NN gt N N N A x Ki E sv Export zoom 300 f Polyine opacity 50 f T RotateLabels 30 P I Meantine Y Polyines Histograms I Left I Right Histogram Bins 3 Options 73 records 73 used 0 selected 52 dmensions Figure 3 17 A parallel coordinates display at a zoom level of 400 With zooming users can more easily discern polylines drawn close to each other Lines are infinitely thin by definition that is they can always be drawn with a thickness of 1 pixel irrespective of the zoom level This helps in detecting patterns in otherwise overly crowded areas of the parallel coordinates visualisation See also Section 3 2 for discussion of te
105. enumeration element in both dimensions See Figure 3 7 for an exemplary depiction of parallel sets 20 3 Parallel Coordinates Austria Austria Italy Italy Male UK UK Female a An individual axis in parallel b Two adjacent axes in parallel sets sets Figure 3 7 Parallel sets Bendix et al 2005 are a variation of the original parallel coordinates idea suitable for data with enumerated dimensions While traditional parallel coordinates visualise continuous data such as real numbers enumerated like Gender or Country pose different problems In a parallel sets visualisation instead of points pertaining to individual records sets of enumeration values are connected on parallel axes by polygons a shows how an axis is organised in parallel sets by drawing a rectangle for each element in the enumeration its height depending on the number of records belonging to the element in that dimension b shows how adjacent axes are connected by polygons in red In this fictional example the visualisation shows more records with people from the UK than from Austria or Italy men are from all three countries while there is no record for an Austrian woman The image was adapted from Bendix et al 2005 Parallel sets differ from the approach discussed in Andrienko and Andrienko 2004 which also discusses subset visualisation in parallel coordinates though still based on defining subsets by dividing continuous numeric types int
106. er information The use of curves instead of lines for parallel coordinates is addressed in a different context by Gra ham and Kennedy 2003 see Section 3 3 3 e Randomness can be exploited for reducing clutter too Dix and Ellis 2002 introduce the idea of visualising random samples of the input data set and apply the idea to parallel coordinates in Ellis and Dix 2006 Ease of use is stated as a distinct advantage of such techniques Doubts about the validity of random sampling in data analysis are met with references to other areas in computer science where randomness plays an important role such as keys in cryptography or neuronal networks Johansson et al 2004 approach the problem with neural networks and propose a solution in volving the classification of records using the Self Organising Map SOM algorithm Kohonen 1997 The clusters displayed are then based on that classification Splatting of parallel coordinates polylines is discussed by Zhou et al 2009 The idea here is to exploit time to visualise the process of clustering step by step as each line is added separately to 18 3 Parallel Coordinates x cereal view 0 WH File View Brushes Tools Window Help a A zx zoomy 100 zooor 2108 Gy CR 1 N pal d b gal b a pot yM cp Wa 1765 6 47 5 785 341 5 14 92 25 47 17 07 352 1105 237 1 74 ex pz szal de 35 0 53 0 485 215 0 92 2 47 207 2 105 1 37 1 34 En P
107. ers on dimensions but also set filters for enumera tions In a a range filter is turned into a set filter b shows the graphical representation of a set filter filled circles are drawn for allowed values of the enumeration empty circles are drawn for disallowed ones The circles can be clicked to toggle the state Figure A 8 Histograms can be turned on and off in MDE s parallel coordinates visualisation They can be shown either left of each axis right of each axis or on both sides The colours and opacity of histogram bins as well as the number of bins to be displayed on each axis can be changed by the user A 3 Scatter Plot View 101 A E Y Figure A 9 In MDE s parallel coordinates visualisation polylines are selected by dragging a rectangle with the mouse All records belonging to one of the polyline segments under that rectangle change their state and become selected which is visualised by drawing them in blue This example also shows how unused records are not included in the selection regardless of whether they are under the selection rectangle or not NAN Helsinki N N NN 4 AN Wess S K Figure A 10 When the user moves the mouse pointer over a polyline in MDE s parallel coordinates visualisation that line is displayed in red With some delay the default being 1 second tooltips showing the corresponding record s individual values appear on each axis and thus a group of records can be sele
108. ersection feature selecting records is based on the concept of situations A situation is a named selection composed of constraints defined on each dimension The way a constraint is defined depends on the nature of the dimension the type of which can either be numerical or an enumeration of strings Constraints for numerical dimensions are defined in terms of lower and upper bounds while constraints for enumerations are defined as a subset of allowed values The graphical user interface however provides no way of defining situations they have to be defined manually in the tool s input text files Besides the parallel coordinates visualisation there is a list of all currently defined situations Each entry in the list can be activated separately which results in the situation and all the records it encom passes being visualised using the colour defined for that situation The user may then expand or narrow the situation for axes with real values and toggle allowed values for axes with enumerations Where situations cross colours are mixed automatically the tool s main purpose being conflict analysis 3 5 6 Picviz Picviz is an open source C library originally made by Tricaud et al 2011 and implemented itself with the Cairo 2010 graphics library The developers also provide ready to use command line interface CLI and graphical user interface GUI frontends At the time of writing the project is still in its beta phase Figu
109. es 55 63 and 110 Bostock Michael and Jeffrey Heer 2009 Protovis A Graphical Toolkit for Visualization IEEE Transactions on Visualization and Computer Graphics 15 pages 1121 1128 ISSN 1077 2626 doi 10 1109 TVCG 2009 174 http vis stanford edu protovis Cited on pages 47 48 and 81 Bray Tim Jean Paoli Eve Maler Francois Yergeau and C M Sperberg McQueen 2008 Extensible Markup Language XML 1 0 Fifth Edition W3C recommendation W3C http www w3 org TR 2008 REC xm1 20081126 Cited on pages 55 and 64 Brodbeck Dominique and Luc Girardin 2003 Design Study Using Multiple Coordinated Views to Analyze Geo referenced High dimensional Datasets In Proc 1 International Conference on Co ordinated and Multiple Views in Exploratory Visualization CMV 2003 doi 10 1109 CMV 2003 1215008 http www macrofocus com public products infoscope Cited on pages 9 11 12 15 22 30 33 and 80 Brody Howard Michael Russell Rip Peter Vinten Johansen Nigel Paneth and Stephen Rachman 2000 Map making and myth making in Broad Street the London cholera epidemic 1854 The Lancet 356 9223 pages 64 68 doi 10 1016 50140 6736 00 02442 9 http www ph ucla edu epi snow mapmyth mapmyth html Cited on page 4 Buschmann Frank Regine Meunier Hans Rohnert Peter Sommerlad and Michael Stal 1996 Pattern Oriented Software Architecture Volume 1 A System of Patterns Wiley ISBN 0471958697 Cited on page 67
110. ference is defined by Mader 2007 as the number of dimensions which is typically significantly higher in vector space data For example Widdows et al 2006 describe a linguistic vector space data set with 1000 dimensions In contrast multi dimensional data visualisation is more limited with regards to the num ber of dimensions as explained by Yang et al 2007 However most traditional multidimensional visualization techniques suffer from visual clutter and only scale up to tens of dimensions Up to now few multidimensional visualization systems have claimed to be scalable to data sets with hundreds of dimensions 2 3 Multi Dimensional Data Visualisation 7 30 0 3s G0 2 Alm HOS ss dins dd ioa CO dina Figure 2 3 A matrix of multi dimensional data with n records and m dimensions This is an abstraction of input for information visualisation software tools capable of dealing with multi dimensional data Name Area km Elevation m Population Graz 127 56 393 255304 Milan 183 77 120 1306637 London 1706 8 24 7556900 Figure 2 4 An example of concrete multi dimensional data for a list of cities with three records and four dimensions Name Area Elevation and Population This could be actual input for an information visualisation software tool There is also a subtle semantic difference in that in multi dimensional data the meaning of the dimen sions themselves is equally important Mader 200
111. formance gains in trade for these advantages Performance is one of the major goals in the development of MDE and the fact that C compiles code to native executables accessing the underlying operating system without any intermediate layers providing for closer vicinity to hardware makes the language more than apt for a desktop visualisation application Other than that C is one of the most used programming languages in computer science and software engineering which means that extensive documentation and support from a large community are available MDE was developed with Microsoft Visual C 2010 Express Edition Microsoft 2010b which is available for free on Microsoft s official website The source code package of MDE contains a solution file which developers can use to build MDE with all necessary compiler switches and automatically execute unit tests Doxygen van Heesch 2011 is used for documentation at source code level all necessary means to build documentation in various formats at class and function level are provided Note that MDE was tested only with Microsoft Visual C 2010 Express Edition on Windows XP and Windows 7 Porting MDE to different platforms or different compilers such as GCC Free Software Foundation 2011 is an example of future work However MDE s design principles mentioned above paired with the cross platform compatibility of all 3 party libraries involved inevitably ease any such undertaking significa
112. ges wellcome ac uk indexplus image L0063431 html Cited on page 5 Widdows Dominic Scott Cederberg and Beate Dorow 2006 Visualisation Techniques for Analysing Meaning In Sojka Petr Ivan Kopecek and Karel Pala Editors Text Speech and Dialogue Lecture Notes in Computer Science volume 2448 pages 107 114 Springer doi 10 1007 3 540 46154 X_14 Cited on page 6 Wielemaker Thomas 2010 Information Visualization and Conflict http mastersofmedia hum uva n1 2010 05 25 information visualization and conflict Cited on page 4 Wikisource 2011 Ode to Joy http en wikisource org wiki Ode_to_Joy Cited on page 1 wxWidgets 2010 wxWidgets http www wxwidgets org Cited on pages 54 57 and 110 XmdvTool 2010 XmdvTool http davis wpi edu xmdv Cited on pages 8 12 17 18 44 and 80 Bibliography 129 Yang Jing Daniel Hubball Matthew O Ward Elke A Rundensteiner and William Ribarsky 2007 Value and Relation Display Interactive Visual Exploration of Large Data Sets with Hundreds of Dimensions IEEE Transactions on Visualization and Computer Graphics 13 3 pages 494 507 doi 10 1109 TVCG 2007 1010 Cited on page 6 Zhou Hong Weiwei Cui Huamin Qu Yingcai Wu Xiaoru Yuan and Zhuo Zhuo 2009 Splatting the Lines in Parallel Coordinates Computer Graphics Forum 28 3 pages 759 766 ISSN 1467 8659 doi 10 1111 j 1467 8659 2009 01476 x Cited on page 17 Zhou Hong Xiaoru Yuan Hu
113. gle drawCircle drawRectangle doDrawLine doDrawCircle doDrawRectangle ParallelCoordinatesVisualisation VisualisationImplementationOpenGL doDrawLine doDrawCircle doDrawRectangle Figure 4 8 This class diagram shows the bridging between visualisation frontend and back end in MDE visualisation the frontend stores a pointer to VisualisationImplementation the backend Both classes are abstract derivees of visualisation are concerned with the logic of creating structured graphics VisualisationImplementation is concerned with trans lating that logic into actual on screen rendering ParallelCoordinatesVisualisation and VisualisationImplementationOpenGL are given as examples of concrete subclasses here If for example ParallelCoordinatesVisualisation needs to draw a line then it calls the protected method Visualisation drawLine The base class will delegate the call to VisualisationImplementation drawLine from where it will be dispatched to the concrete rendering backend s visualisationImplementationOpenGL doDrawLine function which fi nally draws a line on the screen It follows that Visualisation derived classes do not need to know how the visualisations they pro duce are implemented Figure 4 8 is a class diagram of the bridge In summary Visualisation is implemented in terms of VisualisationImplementation P e Visualisation s only public method
114. gure 3 12 shows how axes are moved in parallel coordinates visualisations Following up on the previous example users might want to know more about how Medium Rents and Paid Vacation Days Per Year are related so the axis about rents is moved next to that about paid vacation days 3 4 3 Looking for Patterns With dimensions reduced to a suitable subset and arranged in a suitable order the pattern of lines between the axes can be analysed Siirtola and R ih 2006 explain that the crossings of lines should be observed if many lines cross and if they become more parallel to each other after having flipped one of the axes upside down a correlation between the two dimensions can be deduced Figure 3 13 shows how flipping one of the axes immediately reveals that 1 Rents and paid vacation days are not much related to each other 2 People living in cities with low rents tend to work longer to earn enough money to buy 1 kg of bread 3 4 4 Selecting Records Brushing Selection means choosing a subset of records of particular interest a selection is usually highlighted by colouring the records differently Figure 3 14 continues the previous example It shows how one of the lines stands out between two apparently correlated dimensions This record has been selected and is highlighted in blue Its details are shown and reveal the city of Caracas to be an example of a place where rents are much higher compared to other cities in relation to the tim
115. he FTGLTextureFont Class An identity matrix is first positioned such that the desired text coordi nates correspond to its point of origin It is then rotated Finally text is drawn at position 0 0 of the transformation matrix visualised here as the rotated rectangle around the text The operation is completed by restoring the previous transformation matrix 89 90 5 Selected Details of the Implementation Chapter 6 Conclusion and Future Work To them and their posterity will we commit our future They will continue the voyages we have begun and journey to all the undiscovered countries boldly going where no man where no one has gone before James T Kirk in Star Trek VI The Undiscovered Country 1991 This thesis began with an introduction to information visualisation in Chapter 2 It placed information visualisation in a historical context and demonstrated that the concept existed long before computers were invented Furthermore it defined the term information visualisation to distinguish it from general data visualisation Geographical visualisation was given as a prime example of visualisation often falsely attributed to the field of information visualisation Having defined the precise scientific meaning of multi dimensional information visualisation a num ber of popular techniques were introduced scatter plots histograms star plots and similarity maps The growing importance of user interaction
116. he globe UBS 2009 introduced in Section 2 3 1 again using the MD VS software application developed by Mader 2007 showing similar cities corresponding to the peak of an island 2 3 Multi Dimensional Data Visualisation InteractiveScatterplot 1 cities2009 csv 2 o xj zom A PA id J Vectornumbers J VectorCluster Pos 431 397 0 8774999976158142 0 7924999594688416 Figure 2 5 MD VS scatter plot developed by Mader 2007 visualising a data set about prices and earnings in different cities around the globe provided by UBS 2009 The data set was turned into a comma separated value file suitable for opening in MD VS with the InfoScope tool made by Brodbeck and Girardin 2003 The horizontal axis of the plot corresponds to dimension Gross purchasing power the vertical axis to Net purchasing power The dots represent data points in this case individual cities Their position on the plot therefore depend on their values in the two dimensions corresponding to the axes Dot size is a third information channel the greater the radius of a dot the greater the data point s value in this case in the dimension Prices excluding rent 10 2 Information Visualisation Popularity lt 9 A n n k i A A a e ee 8 f IT an r Baer T o Roe E Zeven nee a ae T en SE see tern 27 TaT a che st 6 ee ae nn ein En Be ae a ei eee L22 A Sig
117. hen an enumeration dimension is selected IntegerDimensionInformationWindow DimensionTableDataViewSubwindow for when an integer number dimension is selected oDimensionInformationWindow DimensionTableDataViewSubwindow for when no dimension is selected ParallelCoordinatesAxis Means to interact with an axis in ParallelCoordinatesVisualisation ParallelCoordinatesAxisMenu Menu for axes in ParallelCoordinatesVisualisation ParallelCoordinatesFilterCircl Means to interact with set filters in ParallelCoordinatesVisualisation ParallelCoordinatesFilterTriangle Means to interact with ranged filters in ParallelCoordinatesVisualisation ParallelCoordinatesVisualisation InteractiveVisualisation with parallel coordinates ParallelCoordinatesVisualisationFactory Creates ParallelCoordinatesVisualisation instances RealDimensionInformationWindow DimensionTableDataViewSubwindow for when areal number dimension is selected RecordTableDataView DataView with a text table of records RecordTableDataViewFactory Creates RecordTableDataView instances RecordTableDataViewListControl List view control for RecordTableDataView ScatterPlotVisualisation InteractiveVisualisation with a scatter plot ScatterPlotVisualisationFactory Creates ScatterPlotVisualisation instances TableDataViewListControl wxListV
118. hism involved Data is stored in class Matrix after it has been loaded see the Data Input module discussed in Sec tion 4 2 3 The constructor of Mat rix is templatised in order to achieve complete flexibility with any data sources Filters on dimensions are maintained by Holder which stores a collection of DimensionFilter instances separately for each dimension in its encapsulated matrix A filter restricts a dimension to a subrange or subset of values This way filters determine which records are considered used and which ones are considered unused There are two different ways of defining a filter e Using a percentage range Since the normalised complete range of data in a dimensions is always 0 0 1 0 a filter of this type is defined as a pair of double values For example a filter might be 0 1 0 9 which means that any record whose normalised point in that dimension is less than 0 1 or greater than 0 9 will be considered unused e Using a set of allowed values This mode can only be applied to dimensions of type enumeration A DimensionFilter defined as such contains a set of used enumeration indices A record s enumeration index in the corresponding dimension must be an element of that set otherwise it is considered unused The design of this module generally assumes that there will be many records and few dimensions which is why no effort is made to optimise dimension specific data or algorithms whereas such
119. howing the record s data values appear 3 5 3 parvis parvis is an open source Java desktop application developed by Ledermann 2003 It is a general purpose parallel coordinates visualisation tool without any particular application area The text based STF file format is used for data input At the time of writing the project is discontinued Although the latest release version number is only 0 3 1 parvis is a mature and usable piece of software Figure 3 21 shows the tool Dimensions in the display can be reordered freely by horizontally dragging the axes but an axis cannot be repeated or removed from the display Individual axes can be translated and scaled vertically There is however no way to zoom the entire display When the tooltips option is activated records are focussed when moving over them with the mouse pointer and the record s data values appear as tooltips over the data points on the axes Selecting the records however is a rather complex operation because the only selection type supported by parvis is angular brushing Records cannot be selected by a simple single click by creating an axis interval or by drawing a rectangle 3 5 Software Applications 37 BPrarvis cameras stf Pale Fa File Edit View Help br Order file 4C4 Programmelpanis cameras stf Load File tla Scale ls Translate Ma Brush Edit Mode Datasource 1 1 11 554734 j d 0 00
120. ider or the check boxes for mean line and histograms Visualisation then starts the process of rebuilding its visualisation or parts of it by sending the custom wxEVT_VISUALISATION_MUST_BE_REBUILT_AFTER_DATA_CHANGE event to itself and handling it in function onVisualisat ionMustBeRebuiltAfterDataChange The nature of the change dictates how visualisation proceeds The data Change and options Change objects are inspected to decide on further action e If the change is more comprehensive than just a change of the record currently being focussed then rebuildVisualisation is called Subsequently in VisualisationImplementationOpenGL doClear the pixels cache is cleared The background cache remains for now untouched Visualisation commences wxEVT_IDLE event handling which means onIdle is called repeat edly until all records of the input data set have been processed This design keeps the window responsive while extremely comprehensive data sets are being visualised 5 1 Multi Phase Caching 078840291 0 966428 098165987 0 99043326 0 99521884 0 96273723 096119518 0 95457247 0 94112327 094335933 10 8640254 o 033748985 0 000509042 01037386217 0 096224911 0 052951044 0025382179 01145557 0 008026362 0065965903 0 071740001 0 002782000 united untited2 untitled united untiteds untitled united untitled untitleds untitled 10 untied a Background 078840291 0 9a966428 098165987 0 999433
121. ied The resulting code shows that there is no direct mapping between VisualisationElement objects and SVG elements the fill colour is changed twice which results in individual visualisationElement objects whereas in the SVG colours become attributes of corresponding rect tags 4 2 Modules 75 Name Purpose Colour RGB colour code Exception Base exception for all errors related to presentation structure InvalidColourException Exception when an invalid Colour is to be created Rectangle Geometrical rectangle definition Text Geometrical text definition VisualisationElement Element of a VisualisationStructure VisualisationStructure Collects VisualisationElement instances VisualisationStructure Geometricallterator Iterates over the geometrical elements of VisualisationStructure VisualisationStructure Iterator Iterates over the elements of VisualisationStructure boundsOutsideOfRectangle Whether an element is entirely outside of a rectangle buildSvg SVG code from a VisualisationElement range nclosedByRectangle Whether an element is entirely within a rectangle equal Floating point aware equality comparison greaterOrEqual Floating point ware greater than or equal to comparison hslToRgb Converts from HSL to RGB intersectsRectangle Checks whether a VisualisationElement inter
122. iew is the ability to show only used or only used and selected records This feature is enabled by clicking on the radio buttons on top of the window An example of using it is to click on Show only used and then set filters on A 4 Records View 103 cities2009 csv Multi Dimensional Explorer ax t t 4 4 t t 4 1255 1198 76700 72300 112200 112000 93600 93700 58300 43000 48500 1800 1200 1800 Cities Gross purchaNet purchasirPrices excl A credit Secretary USSaleswoman Female indusF emale seo 20on 100 poa MENE Feen BR gt 7 Axis Bank redit derk USD per year y Options 73 records 44 used 27 selected 52 dimensions Figure A 12 MDE synchronises the states of records across different visualisation windows This example shows a parallel coordinates view and a scatter plot view opened at the same time Records which have become unused by setting a filter in the parallel coordinates window are displayed using grey polyline in the parallel coordinates window and using grey dots in the scatter plot window as can be seen in the right part of the window The user draws draws a rectangle in the latter selecting all records whose dots are under it this turns the dots blue Corresponding polylines in the parallel coordinates are now displayed in blue as well Record 5 Barcelona 65 7 73 6 83 5 65 3 54 9 61 5 1775 23 0 21 37800 29700 21800 21200 28000 73 re
123. iew with sorting by column used in RecordTableDataView Table 4 7 Classes in MDE s Presentation module used for concrete data views rather than the general mechanisms of the module 74 4 Multi Dimensional Explorer setFillColonr structure Corrour 2557 0 drawRectangle 0 0 0 0 0 3 1 0 setFiliColounr structure Colour 25 07 0 drawRectangle 0 6 0 0 1 0 1 0 Y VisualisationStructure Y std vector lt VisualisationElement gt gt VisualisationElement Different members for different types of visualisation elements lt xml version 1 0 standalone no gt lt DOCTYPE svg PUBLIC W3C DTD SVG 1 1 EN http www w3 org Graphics SVG 1 1 DTD svg11 dtd gt lt svg version 1 1 width 600 height 400 xmlns http www w3 org 2000 svg gt lt rect x 0 y 0 width 200 height 400 fill 00FF00 gt lt rect x 400 y 0 width 200 height 400 fill FF0000 gt lt svg gt Figure 4 9 Class visualisationstructure maintains the logical structure of visualisations Visualisation contains an instance of it and fills it in operations called by subclasses In this example a subclass draws two rectangles of different colour visualisationElement instances are created which are queried later by buildsvg to create SVG code All elements are de fined in normalised coordinates Only when the SVG is created are the absolute width and height of the image specif
124. ine a set of allowed values A circle is drawn for each value of the enumeration Allowed values are represented by filled circles Circles can be clicked to toggle whether that value is allowed or disallowed Figure A 7 shows what a Set filter for the Cities dimension looks like MDE s parallel coordinates view features histograms as shown in Figure A 8 Histograms can be turned on and off individually for both sides of an axis by clicking on the respective check box at the bottom of the visualisation window They represent the number of records in individual subranges of an axis the number of bins to be displayed on an axis is changed via the Histogram Bins spin control Presumably the most common user interaction feature of MDE s parallel coordinates3 is the selec tion of records When a polyline is clicked the record it belongs to is selected A group of polylines 98 a One four dimensional record in parallel coordinates A MDE User Guide b Two four dimensional records in paral lel coordinates Cities Gross purchasing power Net purchasing power Prices SVG Export Zoom 100 Polyine Opacty 50 f I RotateLabels 1 I Meantine IV Polyines Histograms Left Right Histogram Bins 5 4 Loss c Parallel coordinates in MDE Figure A 2 The concept of parallel coordinates is to arrange axes i
125. ing Colour shading is an operation in which one dimension is selected by the user and every polyline is then assigned a colour within a colour range according to the record s data value in that dimension Figure 3 16 is an example of colour shading the operation is applied to the first dimension colours ranging from black over red to white The higher a record s data value in the first dimension the darker its polyline This aids in detecting correlations to other dimensions One can for instance immediately observe that the order of the records in the dimensions on the right is largely the same as in the first 30 3 Parallel Coordinates Figure 3 16 Colour shading is a parallel coordinates operation in which one dimension is selected by the user and every polyline is then assigned a colour within a colour range according to the record s data value in that dimension In this example colour shading is performed on the first dimension and the colour range is defined as black red white The higher a record s data value in the first dimension the darker its polyline This aids in spotting correlations between dimensions Here for example the polylines colour distribution on the right largely match that of the first dimension so a user can deduce that records with higher values in the first dimension tend to have higher values in the last few dimensions as well The image was adapted from InfoScope Brodbeck and Girardin 2003 di
126. ing available everywhere they do not have any way to access graphics hardware directly like MDE with its OpenGL backend consequently the software performs more slowly than MDE 82 4 Multi Dimensional Explorer Chapter 5 Selected Details of the Implementation Hic sunt dracones Here be dragons Latin phrase labelling dangerous unknown territories on medieval maps Chapter 4 discusses the Multi Dimensional Explorer MDE software application from a software engineer s point of view This chapter describes two areas of the software architecture which deserve special attention due to their complexity Section 5 1 discusses the multi layered caching mechanism of MDE Section 5 2 explains how MDE manages to embed text in graphics The context of this chapter is the Presentation module In order to increase legibility class names are not prefixed by presentation For example Visualisation refers to presentation Visualisation 5 1 Multi Phase Caching The visualisation and VisualisationImplementation classes of MDE s Presentation module are re sponsible for drawing graphics in visualisation windows As visualisations in MDE are dynamic they change often whenever the user modifies some parameters of the visualisation or uses its interaction fea tures Recreating the entire visualisation from scratch every time a minimal change occurs is not feasible This is particularly true in the case of ParallelCoordinatesVisual
127. is its constructor and is used by outside code to place visu alisations into the application s general user interface e Visualisation s protected methods are all non virtual and are used by derived classes to build concrete visualisation schemes e Visualisation s virtual private methods are to be implemented by derived classes which con stitutes an application of the Template Method design pattern by Gamma et al 1994 Those implementations are implemented in terms of the protected base class methods mentioned in the previous item e Visualisation s non virtual private methods are helper functions and wxWidgets event handlers Visualisation contains a breach of object oriented pureness It knows of the existence of its spe cial base class InteractiveVisualisation and internally downcasts the this pointer to Interactive Visualisation in order to be able to call certain interaction related functions Those functions should 4 2 Modules 71 not be used by anyone other than visualisation which is why they are private and Interactive Visualisation declares Visualisation a friend in order to grant access to them Visualisation declares Interactive Visualisation a friend too so that InteractiveVisualisation may access private functions of visualisation not meant to be used further down in the inheritance hierarchy This breach of object oriented pureness is justified by the fact that the separation into Visualisation and Interac
128. isation the parallel coordinates component of MDE with its large variety of interaction mechanisms The scatter plot visualisation implemented in ScatterPlotVisualisation is much simpler The parallel coordinates visualisation is also more complex than the scatter plot visualisation for the sheer number of graphics primitives which are drawn Visualising N records and M dimensions requires N x M 1 lines to be drawn in parallel coordinates whereas a scatter plot only draws N glyphs dots or other icons for any given number of dimensions For these two reasons ParallelCoordinatesVisualisation Shall serve as the motivating example for this section However MDE is also designed to be independent of any particular visualisation scheme Therefore any optimisation mechanism in its software architecture has to be defined in abstract general terms VisualisationImplementationOpenGL the main rendering engine of MDE implements those mechanisms in OpenGL Shreiner 2009 Khronos Group 201 1a and is used as an example throughout the rest of this section VisualisationImplementationGraphicsRenderer 1s not considered Optimisation of MDE s rendering performance is built on the idea of caching the pixels of a visual isation after pre defined steps Concrete subclasses of Visualisation know about those steps and can structure the creation of their visualisations accordingly Three such steps are identified 83 84 5 Selected Details of the Imple
129. isualisation might sound like a very general term but has a precise scientific meaning which must be established before discussing any material in detail In particular geographical maps do not usually belong to that area The chapter then discusses the importance of user interaction in today s visualisation software and aspects of multi dimensional information visualisation scatter plots star plots histograms and VisIslands are introduced Chapter 3 explains the mathematical background of parallel coordinates The basic idea is to arrange axes parallel to each other on a plane rather than aligning them in an orthogonal way The individual data points are then connected by polylines across all axes This way more than three dimensions can be displayed at the same time However this visualisation scheme can also be problematic when large data sets must be visualised because the display soon becomes cluttered by many lines The thesis discusses the many techniques proposed to handle this problem by means of clustering It explains also a select number of experimental approaches to extend the original parallel coordinates idea such as 3D visuali sations or replacing polylines with curves The chapter is concluded with an explanation of how a user interacts with parallel coordinates to analyse data A large number of software applications implementing parallel coordinates are reviewed and their individual strengths and weaknesses are outlined In m
130. iversity of Maryland Human Computer Interaction Laboratory http www cs umd edu projects hcil Cited on page 10 Heinrich Julian and Daniel Weiskopf 2009 Continuous Parallel Coordinates IEEE Transactions on Visualization and Computer Graphics 15 6 pages 1531 1538 ISSN 1077 2626 doi 10 1109 TVCG 2009 131 Cited on page 22 Holten Danny and Jarke J van Wijk 2010 Evaluation of Cluster Identification Performance for Dif ferent PCP Variants Computer Graphics Forum 29 3 pages 793 802 doi 10 1111 j 1467 8659 2009 01666 x http www win tue nl dholten papers pcps_eurovis pdf Cited on page 17 Inselberg Alfred 1985 The Plane with Parallel Coordinates The Visual Computer 1 2 pages 69 91 doi 10 1007 BF01898350 Cited on page 13 Inselberg Alfred 2007 Parallel Coordinates Visualization Exploration and Classification of High Dimensional Data chapter 14 pages 643 680 Number III in Springer Handbooks of Computational Statistics Springer ISBN 3540330364 doi 10 1007 978 3 540 33037 0_25 http gap stat sinica edu tw HBCSC Cited on page 13 Inselberg Alfred 2009 Parallel Coordinates Visual Multidimensional Geometry and Its Applications Springer ISBN 0387215077 Cited on pages 13 26 38 and 39 Inselberg Alfred 2010a Parallax Software for Multidimensional Visualization and Automatic Clas sification http www kdnuggets com software parallax html Cited on pages xi 38 and 80
131. kthrough of the application explaining its various features Figure 3 22 shows Parallax For file input Parallax uses simple text files with the default file name ending par In these files records are separated by line breaks while elements within a record are separated by spaces There are only numerical values no enumerations or strings One can however denote undefined null values in the input data set In the visualisation these values appear below the bottom of their corresponding axes Parallax allows users also to create new DAT files In order to do so the current selection is isolated which is to say that all records currently not selected by any query are removed from the data set The remaining records may then be saved as a new file 3 5 Software Applications 39 Axes can be reordered and removed freely but any axis may appear only once The application however automatically generates a list of all meaningful axis permutations to choose from based on the observation in Inselberg 2009 pp 389 390 that few permutations are needed to obtain all possible adjacent pairs of axes In addition to this convenience feature Parallax also features vertical flipping of axes Parallax has scatter plot and histogram visualisations In the scatter plot display record selections are synchronised with the parallel coordinates display It is also possible to select records from within the scatter plot display by drawing a multi sid
132. lected The other black ones are not selected The idea for this concept and the source of this adapted image are from Hauser et al 2002 it works similar to how items are selected in everyday file managers like Windows Explorer Microsoft 2011 or the Finder in Mac OS X Shadovitz 2007 Furthermore if additional visualisations of the same data set are present see Section 2 3 selections should be synchronised in both directions Records selected in a parallel coordinates visualisation should be highlighted as selected in any other views while records selected in the other views should likewise be highlighted as selected in the parallel coordinates view For example if a user selects dots corresponding to records in a synchronised scatter plot window which may also be achieved by enclosing them with a box the polylines corresponding to the same records should be highlighted in the application s parallel coordinates window Angular brushing discussed by Hauser et al 2002 is the name of a special parallel coordinates record selection technique which considers the angle of lines between two axes The degree of the angle of a line reaching from one axis to the next is in the range 0 180 The brushing angle can likewise be defined as an interval of degrees This is especially useful for discerning records whose values go against the general trend between two axes Figure 3 15 shows an example of the concept 3 4 5 Colour Shad
133. lisation a group of axes can be selected by drag ging a rectangle with the mouse in the display area above or below the polylines The user can then right click on any of the axes to open context menu with which all selected axes can be hidden at once 115 8 11 120 14 5 ross purchas jower Go p A Ss Pen Lire a a Click b Moving an able Axis Parts of an Axis Figure A 5 A parallel coordinates axis in MDE presents a number of interaction features to the user Clickable parts of an axis are shown in a The arrow on top 1 can be clicked to toggle flipping The triangles on top and bottom can be dragged to define filtered ranges on the poly lines 2 displays the original triangle position 3 is the filter handle currently being dragged the upper limit of the range is shown above it too 4 is the bottom handle of the range filter b shows an axis being dragged horizontally with the mouse The minimum and maximum value of the dimension are shown at the top and bottom of each axis as well 100 A MDE User Guide 4 y 100 0 12 0 T T Figure A 6 A parallel coordinates axis in MDE can be flipped upside down by clicking on the arrow symbol above each axis The arrow direction changes as well x Range Maximum allowed value Amsterdam zj Minimum allowed value earch y a Editing a Filter b Circles of a Set Filter Figure A 7 MDE supports not only range filt
134. llow me to write Visual Basic code on his old 386 machine and eventually when I had enrolled at university he would share with me important books like Effective C the teachings contained therein ever present in the writing of the software for this thesis As far as programming skills are concerned I also owe a lot to the regulars of the Usenet newsgroup comp lang c I gained many insights by just silently following their often heated discussions about the C programming language I m very grateful also for Alfred Inselberg allowing me to evaluate his Parallax software for free Ross Shannon and Tom Holland for providing me with a copy of their Paired PCV tool Daniel Keim for the source code of VisDB the folks in de comp text tex for help on TEX issues and Anita for her moral support Further help came from my esteemed colleague and good old friend Stefan His help comprised useful IAIRX techniques for generating class diagrams and very many relaxing coffee breaks every one of them sorely needed My supervisor Keith Andrews provided the IAIEX skeleton for my thesis he gave me many valuable hints which really improved my work and he sacrificed a lot of time for our weekly meetings He even gave me one of his own notebooks to develop my software He used to be my teacher my boss and my Bachelor s thesis supervisor He guided me through my studies at Graz University of Technology and I think it is very fitting that their c
135. looking at the visuali sation of a large data set should first obtain an idea of the entirety of the data available overview then a closer view of the data points in a particular area of interest zoom then be able to remove from the view those data points which are not relevant to the task at hand filter and finally be shown all information available for particular data points by explicit request details on demand With information visualisation becoming ever more complex and more interactive usability becomes a greater concern too Andrews 2006 2008a discusses how traditional usability evaluation methods are inappropriate for testing new visualisation techniques Cawthon and Moere 2007 describe a case study exploring the relationship between visually pleasing data representation and the perceived usability of a visualisation Keeping a software application responsive during user interaction contributes greatly to usability as well Piringer et al 2009 discuss the use of multi threading to improve information visualisation software in this regard They observe the problem of reduced responsiveness when software performance suffers from the amount of data to be visualised The crux of the matter is that the time needed to perform a particular rendering step may exceed the time span between two consecutive user decisions meaning that the software will not respond in time before the second action a mouse click for example To mitigat
136. m mers to draw text specifying a font file from which to read character shape definitions FTGL supports TrueType fonts Apple 2002 allows any geometrical operations to be applied to rendered text and al ternatively supports a less flexible but faster type of text rendering Geometrical operations are needed for rotated axis labels in MDE s parallel coordinates visualisation the faster but less flexible type of rendering is used to display non rotated axis labels The complexity arising with MDE s use of the FTGL library is two folded First some way to map from a font s name that which a user sees in the end user interface to the path of its file in the operating system s file system is required Second the font file must be integrated into the OpenGL visualisation with the library s Application Programming Interface API These two aspects are discussed in the following subsections Parallel Coordinates 5 2 Rendering TrueType Fonts in OpenGL E General Parallel Coordinates Scatter Plot Records F Font Of Min Max Values IV Font Of Axis Names I Font Of Tooltips T Font Size Of Axis Names In Pixels T Font Size Of Min Max Values In Pixels I Font Size Of Tooltips In Pixels T Colour OF Axis Names Colour Of Min Max Values T Colour Of Tooltips T Tooltip Shadow Colour Mininum Maximum Values Axis Label Style arial AngsanaUPC Bold Italic JAngsanaUPC Italic Another Typewriter Aparajta Aparajita B
137. masters_thesis Cited on page 3 Bibliography 125 Lex Alexander Marc Streit Ernst Kruijff and Dieter Schmalstieg 2010 Caleydo Design and Evalua tion of a Visual Analysis Framework for Gene Expression Data in its Biological Context 57 64 pages doi 10 1109 PACTFICVIS 2010 5429609 http caleydo icg tugraz at publications html Cited on pages 46 and 81 Llopis Noel and Charles Nicholson 2011 UnitTest http unittest cpp sourceforge net Cited on pages 55 76 and 110 Macrofocus 2010 InfoScope Distiller http www macrofocus com public products infoscope tutorial distiller Cited on page 33 Maddock Henry 2008 FTGL User Guide http ftgl sourceforge net docs html Cited on pages 55 86 and 110 Mader Helmut 2007 Visualizing Multidimensional Metadata Development of the Visualization Framework MD VS Master s Thesis Graz University of Technology Austria http www iicm tu graz ac at iicm_thesis hmader pdf Cited on pages 6 7 8 9 12 19 and 33 Meyers Scott 1996 More Effective C 35 New Ways to Improve Your Programs and Designs Addison Wesley Professional ISBN 020163371X http www aristeia com books html Cited on page 53 Meyers Scott 1997 Effective C 50 Specific Ways to Improve Your Programs and Design Second Edition Addison Wesley Professional ISBN 0201924889 http www aristeia com books html Cited on pages 53 and 67 Microsoft 2010a About the
138. mension records with higher values in the first dimension tend to have higher values in the last few dimensions as well Colour shading is similar to selecting records The difference is that selections result from the user s interest in a particular record or group of records whereas colour shading is useful when the user is interested in a dimension as a whole and how that dimension generally relates to other dimensions 3 4 6 Zooming Bauer et al name zooming as the final step in using parallel coordinates for data analysis Indeed as parallel coordinates polylines are infinitely thin by definition they can always be rendered with a thickness of exactly 1 pixel regardless of the zoom level of the entire visualisation This can help discern polylines placed very close to each other Figure 3 17 shows the zoomed region of a parallel coordinates visualisation However if users just need a quick way to identify individual records in crowded areas of the display zooming may not be a desirable solution because it requires users to temporarily sacrifice their complete overview of the data and to constantly zoom in and out This is a rather tedious way to follow a polyline along all axes Temporarily selecting a record and then deselecting it again just to see its entire polyline in a crowded display mixes different tasks and concerns A better approach to provide the user with a record s details on demand is to temporarily focus it wh
139. mentation 1 Rendering the background of a visualisation The background includes elements which are likely to be unaffected by the most common forms of user interaction In the case of parallel coordinates axis labels minimum and maximum value labels as well as flipping arrows on top of each axis are considered the visualisation s background These elements do not change often 2 Drawing the main contents of a visualisation on top of its background This is when graphical elements for individual records are drawn In the case of parallel coordinates polylines for unused used and selected records as well as the mean line histograms and filter handles These elements change more often than the background as record states are modified by the user selecting records and adjusting filters on dimensions 3 Drawing the overlay of a visualisation on top of its background and main contents This is for temporary changes of a visualisation as the user moves the mouse pointer over it In the case of parallel coordinates focussing of records and displaying tooltips for them as well as drawing highlighted axes elements and filter handles and displaying a semi transparent copy of an axis while it is being dragged with the mouse Figure 5 1 gives a better idea of how these steps translate to visualisation phases Unused records are part of the main contents rather than the background due to the observation that the manipulation of filters is a freque
140. mentation The implementations Of ParallelCoordinates Visualisation and ScatterPlotVisualisation serve as further reference oo Ny AUA Uy 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 B 3 Implementing Visualisations 117 include test_visualisation h include data holder h namespace mde namespace presentation TestVisualisation TestVisualisation DataViewConstructionParameters const amp parameters Visualisation parameters void TestVisualisation partiallyCreateVisualisation int record_index if dataHolder matrix dimensionCount gt 0 if dataHolder isRecordUsed record_index setStrokeColour getOption unused_records_colour else if dataHolder matrix recordState record_index data Selected else setStrokeColour getOption selected_records_colour setStrokeColour getOption records_colour drawLine 0 0 dataHolder matrix pointNormalised data DimensionIndex 0 record_index 1 0 dataHolder matrix pointNormalised data DimensionIndex 0 record_index Listing B 6 A complete example implementation for class TestVisualisation Protected member functions of the base class are used to implement partiallyCreateVisualisation which is called individually for each record dataHolder accesses the input data setStr
141. n sional Visual Exploration using Scatterplot Matrix Navigation IEEE Transactions on Visualization and Computer Graphics 14 6 pages 1539 1148 ISSN 1077 2626 doi 10 1109 TVCG 2008 153 https engineering purdue edu elm projects scatterdice html Cited on page 8 Fanea Elena Sheelagh Carpendale and Tobias Isenberg 2005 An Interactive 3D Integration of Par allel Coordinates and Star Glyphs In Proc 11 IEEE Symposium on Information Visualization In foVis 2005 pages 149 156 IEEE Computer Society ISBN 078039464x doi 10 1109 INFVIS 2005 1532141 http innovis cpsc ucalgary ca Research 3DParallelCoordinates Cited on page 8 FitzPatrick Paul J 1960 Leading British Statisticians of the Nineteenth Century Journal of the American Statistical Association 55 289 pages 38 70 ISSN 0162 1459 Cited on page 3 122 Bibliography Fraumeni JF Jr 1968 Cigarette Smoking and Cancers of the Urinary Tract Geographic Variations in the United States Journal of the National Cancer Institute 41 5 pages 1205 1211 http lib stat cmu edu DASL Datafiles cigcancerdat html Cited on page 22 Free Software Foundation 2011 GCC the GNU Compiler Collection http gcc gnu org Cited on page 54 FreeType 2010 FreeType 2 Overview http www freetype org freetype2 index html Cited on pages 55 and 110 Friendly Michael 1991 Statistical Graphics for Multivariate Data In Proc 16 SAS Users Group In tern
142. n be made the centre axis B C and D arranged around it in a circle Figure 3 8 explains the concept 3 3 Variations 21 I A B C D a An axis in traditional 2D par allel coordinates has at most neighbours D C b The centre axis in 3D paral lel coordinates can have multi ple neighbours Figure 3 8 3D parallel coordinates by Johansson et al 2005a turning parallel coordinates from plane to space by placing one axis into the centre and arranging the others around it in a circle a shows a disadvantage of 2D parallel coordinates in that an axis B in this fictional example can have at most two neighbours even when the user might want to explore the relationship of one dimension to many others b an adaption of images appearing in Johansson et al 2005a shows 3D parallel coordinates overcoming this problem by allowing multiple neighbouring axes All other axes are now neighbours of the central axis B 22 3 Parallel Coordinates Moving axes along the Z dimension is not the only way of turning parallel coordinates into 3D We genkittl et al 1997 propose instead to turn the two dimensional axis lines of traditional parallel coordi nates into planes arranged in 3D space Parallel axes transformed to parallel planes move quite far away from the original idea of parallel coordinates whereas the previously discussed approach feels more like a natural evolution of the concept 3 3 3 Curves Parallel coordinates
143. n information visualisation framework by Lex et al 2010 implemented in Java Oracle 2011 and OpenGL Shreiner 2009 Khronos Group 2011a rooted in medical and genetic science although it can be used for any kind of multi dimensional data Other than parallel coordinates it features a number of additional visualisation methods suitable for its scientific problem domain heat maps and a radial hierarchy Different visualisations can be shown arranged next to each other in a special 3D view called a bucket Figure 3 28 shows Caleydo s parallel coordinates component Axes in the parallel coordinates view can be rearranged hidden and repeated freely Records can be selected individually or grouped via angular brushing 3 5 11 OECD eXplorer OECD eXplorer is a free online tool provided by the Organisation for Economic Co operation and De velopment OECD for visual analysis of regional statistics Organisation for Economic Co operation and Development 2011 Jern 2009 It is implemented in Flash Adobe 2011 The software is made for OECD specific data but custom data can be loaded by users in the form of plain text files This feature is not really made for data completely unrelated to OECD statistics but in practise one can adapt any data set to be read correctly by OECD eXplorer Other than parallel coordinates the software features scatter plot and table lens visualisations There is a data grid which displays records and dimensions i
144. n parallel to each other on a plane a shows a four dimensional record p visualised with a vector of three connected lines at coordinates a b c and d on axes A B C and D respectively This demonstrates how the technique can be used to display records with any number of dimensions as long as there is enough horizontal space In b two four dimensional records are visualised with parallel coor dinates c shows MDE s parallel coordinates component implementing the concept visualising 73 records and 4 dimensions SVG Export Zoom 100 Polyine Opacity 50 H ME 0 H J F Meantine I Polyines Histograms I Left I Right Histogram Bins 5 Options Figure A 3 The controls shown by MDE on the bottom of a parallel coordinates visualisation window allow users to modify certain aspects of the visualisation SVG Export saves the visual isation as an SVG file the Zoom spin control serves to zoom the visualisation polyline opacity can be controlled by both a spin control and the slider next to it the rotation of axis name labels can be chosen or turned off completely mean line polylines and histograms can be individually turned on and off Finally there is an Options button to fine tune further aspects of the parallel coordinates visualisation A 2 Parallel Coordinates View 99 sterd00 115 410 112 125 119 237230 0158 84 65 177113 4 458 7G0 3 N N Figure A 4 In MDE s parallel coordinates visua
145. n tabular form and very suitable for this domain 3 5 Software Applications 47 A RegionGroup Total Population Unemployment Labor Scatter Plot Table Lens Data Grid SE11 Stockholm Sweden 1 949 516 5 21 OWL My wy sko3 Stredne Slovens Slovak Republic 1 350 366 13 08 PL33 Svietokrzyskie Poland 1 275 550 8 79 E a ATS Mi 2 SE22 Sydsverige Sweden 1 351 257 7 43 TR21 Tekirdag Turkey 1 458 114 9 48 m Total Population pp Unemployment Rate Labor Productivity USD PPP 2000 Em My Patent Applications per min pp cators Set Indicators an Histogram Settings 20 887 53 Figure 3 29 OECD eXplorer an online tool provided by the Organisation for Economic Co operation and Development 2011 features parallel coordinates visualisation for regional statis tical data Every record in the data set belongs to a region and there is a geographical map to aid data analysis This example screenshot was made from a visualisation of a European subset of data with the parallel coordinates view showing four of the dimensions total population un employment rate labour productivity and patent applications Histograms are displayed on each axis and can be used to select groups of records Records with low unemployment rates have been selected which shows how they are generally related to regions with low population and high labour productivity
146. nates oaoa aaa 3 5 Software Applications ee 4 Miulti Dimensional Explorer 4 1 Software Architecture ee 4 2 Modules s ese a4 al a e ee Oe we 4 3 Comparison With Other Tools o o e e e 5 Selected Details of the Implementation 3 1 Multi Phase Caching lt s 2 2222 280 RE Benin 5 2 Rendering TrueType Fonts in OpenGL 2 2 CK Come 20000000004 6 Conclusion and Future Work ii iv ix xi 13 13 17 19 22 31 53 53 57 76 83 83 86 91 A MDE User Guide A 1 Loading Data A 2 Parallel Coordinates View A 3 Scatter Plot View A 4 Records View A 5 Dimensions View A 6 Options B MDE Developer Guide B 1 Development Environment Setup 0 02 00 000000004 B 2 Preparing a New Visualisation B 3 Implementing Visualisations Bibliography 95 95 97 101 102 104 104 109 109 111 113 129 List of Figures 2 1 2 2 2 3 2 4 2 5 2 6 2 7 2 8 2 9 3 1 3 2 3 3 3 4 3 5 3 6 3 7 3 8 3 9 3 10 3 11 3 12 3 13 3 14 3 15 3 16 3 17 3 18 3 19 3 20 3 21 3 22 3 23 3 24 Florence Nightingale s Wedge Diagram e e e 4 John Snow s Cholera Map 0000 ee ee 5 Abstract Matrix of Multi Dimensional Data o 7 Concrete Matrix of Multi Dimensional Data o 7 Scatter Plot cionado aca an aa Wa ba da ge
147. ne 2007 Interactive and Dynamic Graphics for Data Analysis With R and GGobi Use R Springer ISBN 0387717617 http www ggobi org Cited on pages 42 and 80 Dix Alan and Geoffrey Ellis 2002 By Chance Enhancing Interaction with Large Data Sets through Statistical Sampling In Proc 6 Working Conference on Advanced Visual Interfaces AVI 2002 pages 167 176 ACM doi 10 1145 1556262 1556289 Cited on page 17 d Ocagne Maurice 1885 Coordonn es paralleles et axiales M thode de transformation g om trique et proc d nouveau de calcul graphique d duits de la consid ration des coordonn es parallelles Gauthier Villars http dlxs2 1library cornell edu cgi t text text idx c math cc math idno 00620001 view toc Cited on pages 13 and 14 Eisenberg J David 2002 SVG Essentials O Reilly ISBN 0596002238 Cited on page 71 Ellis Geoffrey and Alan Dix 2006 Enabling Automatic Clutter Reduction in Parallel Coordinate Plots IEEE Transactions on Visualization and Computer Graphics 12 5 pages 717 724 ISSN 1077 2626 doi 10 1109 TVCG 2006 138 Cited on page 17 Ellis Geoffrey and Alan Dix 2007 A Taxonomy of Clutter Reduction for Information Visualisation IEEE Transactions on Visualization and Computer Graphics 13 6 pages 1216 1223 ISSN 1077 2626 doi 10 1109 TVCG 2007 70535 Cited on page 19 Elmqvist Niklas Pierre Dragicevic and Jean Daniel Fekete 2008 Rolling the Dice Multidime
148. nent VisualisationImplementation Class with which visualisation is implemented VisualisationImplementationFactory Singleton factory class for producing VisualisationImplementation objects VisualisationImplementationGraphics Renderer Standard operating system graphics drawing backend VisualisationImplementationOpenGL OpenGL rendering backend VisualisationMustBeRebuiltEvent Event for when a visualisation must be rebuilt Widget Interactive entity in an InteractiveVisualisation window WidgetGroup Group of Widget instances in the same InteractiveVisualisation window changeDimensionType Changes dimension type with message boxes iterationIndicesAreUnique Checks record iteration indices for uniqueness showDataViewError Displays DataView error message Table 4 6 The general classes structs and free standing functions of MDE s Presentation module 4 2 Modules 73 Name Purpose DimensionInformationWindow DimensionTableDataViewSubwindow for when a dimension is selected DimensionTableDataView DataView in which dimension specific informa tion is shown and manipulated DimensionTableDataViewFactory Creates DimensionTableDataView instances DimensionTableDataViewSubwindow Panel on bottom of DimensionTableDataView EnumerationDimensionInformationWindow DimensionTableDataViewSubwindow for w
149. ns Observer being defined as pure interface classes It is necessary that they never directly or indirectly derive from wxPanel or wxPanel s parent classes to avoid a diamond class hierarchy Meyers 1997 The way MDE separates data from presentation concerns is similar to the widely recognised Model View Controller MVC software architecture pattern however using the definition of Buschmann et al 1996 pp 140 141 MDE lacks a separate controller component and is thus more an application of the Document View pattern As Buschmann et al explain In several GUI platforms window display and event handling are closely interwoven You can combine the responsibilities of the view and the controller from MVC in a single component by sacrificing the exchangeability of controllers The view component of Document View combines the responsibilities of controller and view in MVC and implements the user interface of the system As in MVC loose coupling of the document and view components enables multiple simultaneous synchronized but different views of the same document This description fits MDE s architecture wxWidgets not only draws windows but also provides input event handling the responsibility of the controller in MVC Module Data corresponds to the document component and module Presentation corresponds to the view component Module Presentation contains four concrete presentation subclasses which are
150. ns and items in its main window tool bar and menu as shown in Figure B 4 Clicking on the tool bar and menu opens an empty window as seen in Figure B 5 This is because Test Visualisation contains an empty implementation of partiallyCreateVisualisation The next section explains how to actually display something B 3 Implementing Visualisations At this point Test Visualisation has been created and embedded into MDE s graphical user interface However the visualisation does not yet display anything To make graphics appear its partiallyCreate von AUN kh UY yn HK 114 B MDE Developer Guide ifndef MDE_PRESENTATION_TEST_VISUALISATION_FACTORY_H_ define MDE_PRESENTATION_TEST_VISUALISATION_FACTORY_H_ include data_view_factory h namespace mde namespace presentation class TestVisualisationFactory public DataViewFactory public TestVisualisationFactory private virtual DataView doCreateDataView DataViewConstructionParameters const amp parameters virtual std string getName const hi Hendif Listing B 3 The header file for TestVisualisationFactory The class is derived from DataViewFactory and implements its pure virtual functions doCreateDataView and getName Multi Dimensional Explorer Views Window Help SSS Figure B 4 The new visualisation in TestVisualisation automatically appears in MDE s tool bar and menu vo av rw N ml NV ee eee ee ee o ooann Roane SO
151. nsions looks like with parallel coordinates 3 2 Handling Large Data Sets A B c D A B c D d b c a a One 4D point on b Two 4D points on parallel axes parallel axes Figure 3 2 a shows a 4 dimensional point p represented by a vector of three connected lines at coordinates a b c and d on axes A B C and D respectively This example demonstrates how parallel coordinates can be used to display records with any number of dimensions as long as there is enough horizontal space while orthogonal axes are restricted to 3D because of human visual perception In b two 4D points are displayed using parallel coordinates Each of them is represented by a vector of three connected lines cities2009 csv Multi Dimensional Explorer Parallel Coordinates t zr sna nes Zurich 120 145 309 249 50 61 1582 60 12 8 8 9 20 03 00 136 294 153 180 130 2240 016 131 1 37 Cities Gross puNet purctPrices e Prices inGross weNet wageWorking Vacation Time reqiTime req Time req Time reqiCity brea Inflation inflation Inflation Prices inFood basServices Normal IcHousehoBus or trTrain Taxi L SVGExport Zoom 100 Polyine Opacity 50 f I Rotate Labels 50 I Meantine Y Polyines Histograms T Left 7 Right Histogram Bins EA Options Record 39 Milan 80 6 77 0 78 9 64 4 63 7 60 8 1808 25 0 27 19 21 16 780 2 217 2 038 3 5 76 0 457 540 1160 2700
152. nt Hardy Erik Dahlstrom Ola Anders son Jon Ferraiolo Doug Schepers Dean Jackson Chris Lilley Andrew Emmons Anthony Grasso Cameron McCormack Andreas Neumann Scott Hayman Antoine Quint and Robin Berjon 2008 Scalable Vector Graphics SVG Tiny 1 2 Specification W3C recommendation World Wide Web Consortium http www w3 org TR 2008 REC SVGTiny12 20081222 Cited on pages 47 and 71 Ramos Ernesto and David Donoh 1983 1983 Data Exposition Dataset http stat computing org dataexpo 1983 html Cited on pages 37 48 and 50 Roberts Jonathan C 2007 State of the Art Coordinated amp Multiple Views in Exploratory Visu alization In Proc 5 International Conference on Coordinated amp Multiple Views in Exploratory Visualization CMV 2007 IEEE Computer Society doi 10 1109 CMV 2007 20 http www cs kent ac uk pubs 2007 2559 content pdf Cited on page 7 Rundensteiner Elke A Matthew O Ward Zaixian Xie Qingguang Cui Charudatta V Wad Di Yang and Shiping Huang 2007 Xmdvtool Quality Aware Interactive Data Exploration In Proc 27 ACM SIGMOD International Conference on Management of Data SIGMOD 2007 pages 1109 1112 doi 10 1145 1247480 1247623 Cited on page 44 Schiller Friedrich 1786 An die Freude Thalia 1 2 pages 1 5 http www ub uni bielefeld de diglib aufk1 thalia thalia htm Cited on page 1 Severgnini Beppe 2008 L italiano Lezioni semiserie Rizzoli ISBN 88170
153. nt user action in parallel coordinates if the change of a record s state triggered a rebuild of the entire visualisation then the filter handles would not be responsive enough The drawback of this approach is the less likely but still possible use case when filters are not used much while record selections change often because in this case continuously redrawing the same polylines for unused records is wasteful Accounting to both cases through a more flexible design involving a dynamic number of visualisation phases is an example of future work on MDE see also Chapter 6 As for the implementation of this design visualisationImplementationOpenGL contains two std vector lt unsigned char gt member variables in which the pixel output of the first and second step is cached background stores the background of the visualisation pixels stores its main contents What follows is a documentation of how and in which parts of the source code the caching takes place When a visualisation window is first created or when any changes in data or options occur presentation Visualisation calls prepareForNewData and prepareForNewOption two of its non pure virtual private member functions The default implementations do nothing but derived classes may override them to take custom preparations for rebuilding a visualisation or parts of it For example ParallelCoordinatesVisualisation updates its custom GUI controls such as the polyline opacity sl
154. ntation Visualisation DimensionTableDataView m tonNewData lt doOnNewData onNewOption doOnNewData doOnNew ption doOnNewData doOnNewOption A doOnNewOption ss presentation presentation InteractiveVisualisation RecordTableDataView doOnNewData A y do0nNewOption presentation presentation ParallelCoordinates ScatterPlotVisualisation Visualisation Figure 4 7 The class hierarchy in MDE pertaining to visualisation The central component is the abstract class Dataview in the Presentation module It is a wxPanel specialisation de rived also from observers in both the Data and the Options module so that all Dataview in stances are kept up to date with changes in data or program options RecordTableDataView and DimensionTableDataView implement DataView directly Visualisation extends DataView for data presentation involving the rendering of graphics Interact iveVisualisation still abstract adds mouse and keyboard interaction mechanisms ParallelCoordinatesVisualisation and ScatterPlotVisualisation are concrete subclasses of it In module Application MainStatusBar is yet another implementor of data Observer keeping MDE s status bar up to date See Fig ure 4 8 for the further separation of visualisation into frontend and backend 70 4 Multi Dimensional Explorer VisualisationImplementation Visualisation drawLine drawCircle drawLine drawRectan
155. ntly MDE s source code is divided into seven modules Each module resides in a distinct namespace or sub namespace mde is the parent namespace it contains no direct members of its own All names of classes functions or other C entities are contained in one of the sub namespaces Table 4 1 gives an overview of the modules Modules Data Data Input Presentation Presentation Structure and Options are meant to be built as libraries which are then linked to by Application and Test the two executables MDE depends also on a number of 3 party libraries e MDE s graphical user interface GUD is implemented with the cross platform wx Widgets library version 2 8 11 wxWidgets 2010 4 1 Software Architecture 55 Module Name Purpose C Namespace Application Application specific components mde application such as the help menu or status bar Data Data matrix storage normalisation mde data and access synchronisation Data Input submodule of Reading data from external mde data input Data sources such as files Presentation Data presentation such as visu mde presentation alisation windows and rendering backends Presentation Structure Logical structure of visualisations mde presentation structure submodule of Presenta such as geometrical line and circle tion definitions SVG creation Options Program options management mde options such as storing an internal options t
156. o popular belief it is not true that Snow came up with his theory as a result of analysing the map he drew the map as an illustration afterwards Brody et al 2000 Friendly and Denis 2001 cite many other historical examples of maps diagrams and information visualisation dating the beginning of the modern age of information visualisation to the beginning of the second half of the 20 century 2 1 History Figure 2 2 John Snow s cholera map from 1854 is sometimes misunderstood as an important con tribution to the field of information visualisation when it is actually an example of geographical visualisation The map visualised deaths caused by cholera in different parts of London illustrat ing the theory that a particular water pump in the city had caused the outbreak of the epidemic Popular belief has it that Snow came up with his theory through analysis of the map but in truth the map only illustrated his previously established theory Image copyright of the Wellcome Library 201 1b 6 2 Information Visualisation 2 2 Interaction Rather than just being capable of quickly generating visual output computers also have the ability to ac cept input from their users This makes visualisations interactive because parameters of the visualisation can be immediately updated Shneiderman 1996 famously describes the principle of good interactive visualisation as overview first zoom and filter then details on demand A user
157. o ranges 3 3 2 Three Dimensional Displays Parallel coordinates have traditionally been visualised on a 2D display It is however possible to extend the idea to projecting multi dimensional data on 3D space Johansson et al 2005a propose a technique they call clustered multi relational parallel coordinates to implement 3D parallel coordinates The idea is to take the axes of traditional 2D parallel coordinates and put them into 3D space such that all of them still point in the same direction though their position on the plane is changed One axis is placed in the centre of the display and all other axes are evenly distributed around the centre axis in a circle Polylines connect the centre axis with the outer axes there are no connecting polylines between the outer axes themselves In addition Johansson et al 2005a use clustering just as some traditional 2D visualisations do though this is not directly related to the 3D concept With this approach it is possible to explore relations between dimensions more easily In 2D parallel coordinates one instance of the visualisation pertains to only one combination of dimensions for exam ple in a data set with four dimensions A B C and D to explore the relationship between A and each of the other dimensions requires at least two visualisations because the axis corresponding to A may have only two neighbours not three With 3D parallel coordinates there is no such inherent limit A ca
158. okeColour sets the line colour getOption determines appropriate colours for various record states and drawLine finally draws the line 118 B MDE Developer Guide Bibliography Adobe 2011 Adobe Flash Player http www adobe com de products flashplayer Cited on page 46 Ahlberg Christopher and Ben Shneiderman 1994 Visual Information Seeking Using the FilmFinder In Companion 12 Conference on Human Factors in Computing Systems CHI 94 pages 433 434 ACM ISBN 0897916514 doi 10 1145 259963 260431 Cited on pages 8 and 10 Andrews Keith 2006 Evaluating Information Visualisations In Proc Advanced Visual Interfaces AVI 2006 pages 1 5 BEyond time and errors novel evaLuation methods for Information Visualization BELIV 06 ACM doi 10 1145 1168149 1168151 Cited on page 6 Andrews Keith 2008a Evaluation Comes in Many Guises Position Paper at BEyond time and er rors novel evaLuation methods for Information Visualization BELIV 08 http www dis uniromal it beliv08 pospap andrews pdf Cited on page 6 Andrews Keith 2008b Writing a Thesis Guidelines for Writing a Master s Thesis in Computer Science Graz University of Technology Austria http ftp iicm edu pub keith thesis Cited on page xi Andrews Keith 2010 Information Visualisation Course Notes http courses iicm tugraz at ivis ivis paf Cited on pages 3 4 6 and 22 Andrews Keith Christian Giitl Josef M
159. old Aparajta Italic Arabic Transparent Arabic Transparent Bold Arabic T Arial Black Arial Bold Italic Arial Italic Arial Narrow Arial Narrow Bold Arial Narrow Bold Italic Arial Narrow Italic Arial Rounded MT Bold Arial Unicode MS Ariston ArtBrush Medium Asenine dministrator cmd exe Shortcut 263 gsaz ttf A TIP 150 416 ANTQUABI TIF 149 092 ANTQUAI TIF 222 356 aparaj ttf ara jit 173 596 Aquiline ttf 623 628 arabtype ttf ITE 87 Asenine Wide Axis Label Rotation Angle In Degrees Athletic Regular old audimat Bi Reset All Options a Font Names b Font Files Figure 5 2 No obvious mapping exists between font names and font file paths a shows an option control in MDE displaying all available font names to the user b demonstrates that the corresponding files have very different names which follow no logical scheme For this reason programmers must employ platform specific means of determining the paths of font files 5 2 1 Mapping from Name to File Path Font names as they are displayed to the end user of an application do not correspond at all to file names in the file system For example Arial may correspond to the file path c windows Fonts arial ttf on a Windows installation but for obvious reasons this cannot generally be assumed Unavailability of any such logical mapping between name and path poses a
160. on code is found in the constructor of MainWindow a class derived from wxWidget s wxMDIParentFrame to enable the Multiple Document Interface paradigm Microsoft 2010a The class allocates and stores one instance for each of the data Holder options Tree and options Synchroniser Classes Those are never deleted nor should they in order to avoid order of destruction issues at program termination Note that this does not constitute a memory leak as memory usage is constant Note how the module uses abstract application independent components from Options and integrates them into the GUI using opt ionsDialog and related Application classes This way of interaction between the modules can be described such that opt ionsDialog acts as visual frontend of the Options module Which subwindows can be opened by Mainwindow and consequently the contents of the toolbar and Views menu are determined by querying presentation DataFactoryGroup It returns information about visualisations offered by the application and provides MainWindow with pointers to presentation DataViewFactory When a user wishes to open a visualisation window the factory is used to instantiate concrete presentation DataView subclasses using the Factory Method design pattern by Gamma et al 1994 The subclass instance is then put into a new instance of DataViewMDIChildFrame the base class for MDE subwindows and equipped with references to the centralised instances of data
161. on for differing definition names or types in checkEqual s node comparison DuplicateNodeKeyException Exception for duplicate node keys in options file EnumerationValues Values for enumerated options Exception Base exception for option related errors InvalidDefaultValueException Exception for a Definition s invalid default value InvalidIntegerException Exception for when an integer string in the options file is invalid InvalidValueException Exception for when a value in the options file is invalid for its definition Node Node in Tree NodeKeysDifferException Exception for differing node keys in checkEqual s node comparison NoRoot Exception Exception for when the options file has no root Observer Observes Synchroniser OptionsFileCannotBeOpenedException Exception for when the options file cannot be opened OptionsFileNotFoundException Exception for when the options file cannot be found ParsingException Exception for XML code that cannot be parsed Synchroniser Synchronises options across different Observers ree Manages cascading options TreeStructuresNotEqualException General abstract exception for checkEqual UnknownTypeException Exception for unknown types in the options file ValueForUnknownDefinitionException Exception for val
162. on of parallel coordinates typical user interaction features to a clustered display For example how should MDE s record selection rectangle work with clusters What does it mean to drag filter handles on a dimension when records are clustered What should be highlighted and which tooltip values should appear when the user moves the mouse pointer over a cluster In fact the entire user interaction scheme has to be redefined when clustering is present extensive usability tests 93 should be performed to gain more insights in this area rather than implementing one s own ideas without taking the end user into account Similar considerations hold for colour shading which MDE does not yet implement Colour shading conflicts with record selection because both operations modify the colour of the polylines What exactly do users expect when they select records in a parallel coordinates visualisation with activated colour shading Should the selected polylines just become blue ignoring the colour range Or should the colour of selected records follow a different shading range And how intuitive is an application with such heavy overloading of the colour information channel Again it seems advisable to perform user testing possibly with a prototype to see how this feature should be implemented to aid in exploratory data analysis Parallel coordinates have stood the test of time in the fast moving world of computer science They are certainly her
163. onclusion involves him as well Finally I wish to thank the people from all over Europe whom I met during my unforgettable times in Milan and as a result of my studies abroad because it was ultimately their friendship which gave me the strength to believe in my ability to accomplish this project and to complete my studies in Graz the way I had always hoped I would The encouragement they gave me might have come from distant places but I felt it all the more close to me Thank you Chiara and Jennifer for the conversations and e mail correspondence we had when I was unsure about whether to start this project or not You helped me make the right decision I wonder if this thesis would exist if it was not for you Thank you Mathilde Alessandro and Fabio for believing in me right from the beginning Thanks to each and every one of you who helped me in the project s intense final phase by expressing emotional support in one way or the other Giuseppe Mattia Andrea Katrin Ismael Paula Pietro Despina Wojtek Marina Rafal Ottavio Anneke Abel Leticia Alice I think the list of names could go on forever A person sometimes just needs to hear or read a few encouraging words to carry on the struggle and this is exactly what you ve given me Ce la farai You ll make it I heard that sentence very often and my ultimate determination to finally complete my thesis and to really give my best came also from the wish not to disa
164. ontal line across the full window width is displayed its vertical position depending on the first dimension s data value its colour depending on whether the corresponding record is used black unused grey or selected blue This visualisation does not serve any practical purpose but its simplicity makes it ideal for explaining the programming aspects of adding a new visualisation to MDE These contain the C source files of MDE cpp and n Microsoft Visual Studio project files vcprojx as well as the Windows icon files ico and Windows bitmap files bmp needed by MDE s GUI frontend Subdirectory test also contains simple csv and a number of other plain text files used by MDE s unit tests The root directory contains the Visual Studio solution file mde sin which can be used to quickly create MDE from scratch by opening it in Visual C 2010 Microsoft 2010b and build the solution with one of the Release or Debug configurations Before building MDE with the solution file provided a developer must make sure that the following libraries are installed on his or her system e wxWidgets version 2 8 11 wxWidgets 2010 Boost version 1 43 0 Boost 2010 e UnitTest version 1 4 Llopis and Nicholson 2011 e FTGL version 2 1 3 Maddock 2008 e FreeType 2 FreeType 2010 e TinyXML version 2 5 4 Thomason et al 2010 The version numbers indicate those with which MDE was developed and tested More recent
165. optimisation techniques coming from software engineering Furthermore MDE combines interaction features found in other parallel coordinates visu alisation software creating new usage scenarios Unlike a number of other more academic or experi mental toolkits MDE is meant to be used for real data analysis and has been designed and developed with usability in mind Chapter 5 discussed some particularly complex aspects of MDE s design including the multi level caching mechanism and the difficulties in embedding text into OpenGL visualisations Both are impor tant for a usable parallel coordinates visualisation of course The structure of parallel coordinates makes them apt for caching and text is needed to display axis names or informative tooltips The future of parallel coordinates is hard to predict On the one hand the basic approach of arranging axes next to each other on a plane is going into ever new directions 3D parallel coordinates and curves instead of polylines have been proposed and tested in prototypes On the other hand parallel coordinates are already a useful technique as they are They should be embraced much more by software engineers as well as usability engineers to create robust fast reliable and usable implementations MDE is a starting point for this Although the software performs very fast already many additional techniques are not used and might be tested in future Most importantly MDE does not employ OpenGL di
166. or each axis Let the axes be denoted A and B the coordinates shall be denoted a and b The axes are orthogonal but may also be viewed in parallel to each other as shown in Figure 3 1 The coordinates a and b of p remain the same their positions on the corresponding axes are connected by a line A parallel coordinates visualisation has been created The line connecting the axes represents p in the same way as a two dimensional point represents p using orthogonal axes Once the duality between lines and points has been established the concept can be extended p can have many more dimensions and still be displayed by parallel coordinates as long as there is enough horizontal space There is no geometrical limit to how many axes one can display in parallel to each other This is not true for orthogonal axes which can display only up to three dimensions because of human perception Figure 3 2 shows an example of a 4D point p visualised with parallel coordinates p being represented now by the polyline connecting the axes This way any number of points or records can be visualised at the same time As one can see there are no theoretical limitations on the number of dimensions or records displayed at the same time Limits arise in the practical use of parallel coordinates due to limited screen space and because of overlapping lines as discussed in Section 3 2 Figure 3 3 shows what the visualisation of a data set consisting of 73 records and 24 dime
167. orderingAllAtOnce Tests changing the order of all dimension in data Matrix at once TestDimensionVisibility Tests hiding and showing dimensions TestDimensionWrongType Tests wrong types for data Dimension TestFocusRecord Tests focussing of records tGetSmallestGreatestValues Tests retrieving smallest and greatest values from data Matrix dimensions tHolderAddRemoveFilters I ests adding and removing ta DimensionFilter tO data Holder o estHolderApplyFilters I a ests applying data DimensionFilter in data Holder TestHolderMatrixAssignment Tests basic assignment of data Matrix to data Holder TestMat rixOutOfBounds Tests assertions for illegal indices in a data Matrix TestMatrixPointsAsStrings Tests string representation of data Matrix fields TestMatrixPointsNormalised Tests normalisation of data Matrix fields TestWrongPointStringConverter Tests data PointStringConverter with wrong decimal point character Table 4 9 The unit tests in MDE s Test module testing components of module Data 78 4 Multi Dimensional Explorer Name Purpose TestCsvFileReaderFirstLineCaptions Tests data input CsvFileReader with first line as captions TestCsvFileReaderSimple Tests data input CsvFileReader with simple file TestCsvFileReaderWrongFilename Tests data input CsvFileReader wi
168. oser Vedran Sabol and Wilfried Lackner 2001 Search Result Visualisation With xFIND In Proc 2 4 International Workshop on User Interfaces to Data Intensive Systems UIDIS 2001 pages 50 58 doi 10 1109 UIDIS 2001 929925 Cited on page 8 Andrews Keith and Martin Lessacher 2010 Liquid Diagrams Information Visualisation Gadgets In 14 International Conference on Information Visualisation IV10 pages 104 109 ISSN 1550 6037 doi 10 1109 TV 2010 100 Cited on pages 48 50 and 81 Andrienko Gennady and Natalia Andrienko 2004 Parallel Coordinates for Exploring Properties of Subsets In Proc 2 International Conference on Coordinated amp Multiple Views in Exploratory Visualization CMV 2004 pages 93 104 IEEE Computer Society ISBN 0769521797 doi 10 1109 CMV 2004 13 Cited on page 20 Apache Software Foundation 2011 Apache Subversion http subversion apache org Cited on page 109 Apple 2002 TrueType Reference Manual http developer apple com fonts TTRefMan index html Cited on page 86 Apple 2010 OpenGL Programming Guide for Mac OS X Concurrency and OpenGL http developer apple com library mac documentation GraphicsImaging Conceptual OpenGL MacProgGuide opengl_threading opengl_threading html Cited on page 6 119 120 Bibliography Artero Almir Olivette Maria Cristina Ferreira de Oliveira and Haim Levkowitz 2004 Uncovering Clusters in Crowded Parallel Coordinates Visualizations In
169. ost cases low performance and limited user interaction are problems in parallel coordinates soft ware Based on this finding a new software application called Multi Dimensional Explorer MDE was developed for this thesis Chapter 4 discusses it in detail MDE is a general purpose information visualisation software application although special attention has been given to its parallel coordinates component It was designed and implemented from ground up with high performance and high usability in mind The software architecture of MDE is discussed and its highly modular architecture is explained This allows developers to extend the tool in many different aspects for example by adding new visuali sation techniques or by adding new ways of reading data for input German Original Aus der Wahrheit Feuerspiegel l chelt sie den Forscher an Zu der Tugend steilem H gel leitet sie des Dulders Bahn English translation by Wikisource 2011 2 1 Introduction Chapter 5 of the thesis discusses selected details of MDE s implementation which are particularly complex The first of these selected details is MDE s caching mechanism which divides the process of building a visualisation into different steps and maintains caches of the OpenGL pixel output The second is the solution MDE employs for embedding text in OpenGL visualisations This is non trivial as OpenGL itself does not support text additional libraries have to be used for it
170. ppoint you Christian Hackl Graz Austria March 2011 1X Credits I would like to thank the following individuals and organisations for permission to use their material The thesis was written using Keith Andrews skeleton thesis Andrews 2008b Alfred Inselberg provided a copy of his commercial Parallax software for free Inselberg 2010a Ross Shannon and Tom Holland sent me a copy of their Paired Parallel Coordinates View PPCV tool Shannon et al 2008 Daniel Keim gave me access to the source code of VisDB so that I could compile run and review the program Keim et al 2002 The screenshot of the FilmFinder in Figure 2 6 was used with the kind permission of the University of Maryland see copyright notice below University of Maryland Copyright Notice All works herein are Copyright University of Maryland 1984 1994 all rights reserved We allow fair use of our information provided any and all copyright marks trade marks and author attribution are retained xi xii Chapter 1 Introduction From the fiery mirror of truth She smiles upon the researcher Towards virtue s steep hill She guides the endurer s path Friedrich Schiller Ode to Joy 1786 This Master s thesis is about exploratory data analysis using the parallel coordinates technique for information visualisation The history and general definition of information visualisation are discussed in Chapter 2 Information v
171. problem to MDE because the full path of a font file is needed to draw text with the FTGL library Figure 5 2 shows the difference between font names and font paths wxWidgets is no help in this regard either because although it features many font related com ponents it does not contain functionality for retrieving the path of a font Therefore this is an area where MDE has to resort to platform specific implementations Any such platform specific dependen cies are encapsulated in the implementations of the two functions getAvailableFonts and getFont Directory They are not exposed to other components of MDE The following part of this documen tation assumes a Microsoft Windows platform and availability of the Windows API Microsoft 2010c A font path consists of two parts the directory path and the file name For example in C windows Fonts arial ttf the directory path is c Windows Fonts In order to retrieve it MDE s getFont Directory function uses the Windows API function sHGet SpecialFolderLocation passing the predefined value CSIDL FONTS as an argument getAvailableFonts on the other hand creates statically stores and returns a const reference to a std map in which keys are font names and values are corresponding file names The directory is not contained in the map s values For example key Arial might refer to value arial ttf rather than C Windows Fonts arial ttf The Windows registry has to
172. r of class data Matrix can be seen as the link between the parent and child module The central component of the module is the abstract base class FileReader Inside the Data In put module the only concrete derived class is CsvFileReader an implementation for reading comma separated value CSV files This is because the current version of MDE only supports CSV for input The implementation relies on boost escaped_list_separator from Boost 2010 to break up input lines into tokens Outside of module Data Input in module Application application CsvFileReaderWithDialog Is another derivee Of FileReader It is implemented in terms of CsvFileReader and adds a GUI dialogue the Import Wizard from the end user s point of view This is an application of the Adaptor design pattern by Gamma et al 1994 Figure 4 4 is a class diagram of this part of MDE s architecture crossing module boundaries Table 4 4 lists all classes and functions of the Data Input module 4 2 4 Options This module contains the means to load save and query both options chosen by the user such as colours for selected used or unused records and options saved automatically without the user noticing such as the paths of recently opened files Options is self contained and does not depend on any other MDE 64 4 Multi Dimensional Explorer Name Purpose CannotOpenFileException Exception when a
173. re 3 24 shows the GUI frontend of Picviz Although Picviz can be used as a general purpose parallel coordinates visualisation tool its actual purpose as explained by Tricaud and Saade is network security A tool to convert server logs into the tool s own Picviz Graph Description Language PGDL file format is provided PGDL is a text based graph description language Picviz 2010 The introductory example given by the Picviz developers is to analyse Hypertext Transfer Protocol HTTP server logs Records represent different requests dimensions include information such as a request s Uniform Resource Locator URL Internet Protocol IP address HTTP level request method and the user agent string The authors then go on to show how one can immediately spot particularities in the correlations between request methods and URLs or between IP addresses and user agents Tricaud and Amaducci 2009 The Picviz GUI gives users complete control over the order of dimension axes Axes may also appear 3 5 Software Applications P Situvis_vi_1 CER Situvis Situations GO Phone call as MO demonstrating as MO Meeting with supervisor aa GO sym ma MO Reading ma Time Location Colleagues present Noise level Computer activity Supervisor present Calendar status Phone a Current Version of Situvis Paired Parallel Coordinates an individual s social network extracted from Facebook
174. ree Test Unit tests mde test Table 4 1 Modules of MDE s source code Each of them resides in a distinct namespace or sub namespace of its own mde is the parent namespace it contains no direct members of its own All names of classes functions or other C entities are contained in one of the sub namespaces With the exception of Application and Test all modules are meant to be compiled to libraries Application and Test are meant to be compiled to executables statically linking the libraries created from the other modules Visualisation windows use OpenGL version 1 1 for rendering Shreiner 2009 Khronos Group 201 1a Note that wx Widgets features integrated OpenGL support OpenGL font support is enabled by FTGL version 2 1 3 Maddock 2008 FTGL depends itself on the FreeType 2 library FreeType 2010 Mapping between font names and font file paths is handled in a platform dependent way Therefore text in OpenGL visualisations is currently only enabled in Windows where the Win32 API is used directly to retrieve a list of available fonts and their corresponding file paths On other platforms no text appears See also Chapter 5 TinyXML version 2 5 4 Thomason et al 2010 is used to read MDE s options file in Exten sible Markup Language XML format Bray et al 2008 The Boost libraries version 1 43 0 are used by MDE for a variety of features throughout the entire source code Boost 2010 UnitTest ve
175. rent height are drawn on an axis Each rectangle represents an element of an enumeration set such as all countries on earth for a dimension Country Users are also free to group subsets for example in a Country dimension elements might be grouped by continent so that the visualisation effectively has to deal with only six graphical elements Dimensions of a continuous numerical type may be adapted for use in parallel sets by defining a set of exclusive ranges for example less than 50 greater or equal than 50 greater than 100 The height of a rectangle depends on the number of records belonging to the corresponding enumer ation value in that dimension When axes are displayed next to each other the parallel sets equivalent of parallel coordinates polylines comes into play From each of an axis rectangles a number of polygons extend to the adjacent right axis always assuming that all axes in a parallel sets display are of an enumer ated type The right edge of each polygon is connected to a rectangle on the right axis such that every polygon between two axes N and M represents one of the possible combination of the axes respective dimensions With M different enumeration elements in the left axis and N in the right axis N x M combinations are possible resulting also in a maximum of N x M polygons In other words each of the polygons represents the existence of at least one record belonging to one particular
176. ript code by Jackson 2008 72 4 Multi Dimensional Explorer Name Purpose DataView Base class for the display of data in a window DataViewConstructionParameters Tuple of parameters for DataView s constructor DataViewFactory Creates Dat aView instances DataViewFactoryGroup Group of DataViewFactory instances DimensionFilterDialogue Dialogue for adding or editing a data DimensionFilter Exception Base exception for presentation related errors FontException Exception for font related errors InteractiveVisualisation Visualisation with mouse and keyboard interaction NormalisedMousePosition Mouse coordinates in range 0 0 1 0 NormalisedSelectionRectangl Rectangle with coordinates in range 0 0 1 0 used for group selections in InteractiveVisualisation OpenGLFontException Exception for OpenGL font errors RecordIterationStrategy Defines order by which visualisation iterates records to be rendered RecordIterationUnusedFirst Defines record iteration order such that unused records go first RecordIterationUnusedFirstSelected Last Defines record iteration order such that unused records go first selected last TimedSpinControl wxSpinCtrl emitting events also a few seconds after direct text modification imedTextControl wxTextCtrl emitting timed modification events Visualisation Central visualisation compo
177. rise An exact measure obviously cannot be given as the critical number of records will always depend to a certain degree on the data set The following techniques are specifically tailored towards reducing clutter in parallel coordinates e Fua et al 1999 discuss clustering of records under the name of hierarchical parallel coordi nates The clustered display is achieved by creating an internal tree structure from the data set Each node in the tree corresponds to a certain number of data points and their mean value The level of clustering that is how many lines are displayed in the parallel coordinates view depend on a parameter which can be imagined to horizontally split the tree structure the word cut is used for this in the paper By changing that parameter the user of a visualisation can dynami cally change the degree by which records are joined to clustered polylines Figure 3 6 shows the technique implemented in XmdvTool 2010 Zhou et al 2008 take a different approach Instead of basing any clustering on calculation in the data set itself they base their technique directly on the geometry of polylines which would result otherwise from a normal parallel coordinates visualisation Lines are denoted as belonging to the same cluster according to their parallelism and their position Actual visual clustering is achieved by turning straight lines into curves leaving the colour aspect of the visualisation free to carry oth
178. rrent Jog Choose color amp glyph x coe Ele Options Brush T Persistent Annulla E dan Point brushing ATE x palmtoleic dge brushing x stearic e a x linoleic x linolenic were X eicosenoic sv Scatterplot Pirhi ate Region Region palmitic palmitoleic stearic olve csv 572 x 10 C Programmelggobi datalolive csv Figure 3 25 GGobi an open source GTK application featuring parallel coordinates visualisation The tool offers various other visualisation types and synchronises selections and data manipula tions in each of them Here parallel coordinates are shown with a scatter plot of the first two dimensions GGobi s most distinguishing feature is a set of extensive data manipulation and statistical tools Note that parts of this image have been inverted to increase legibility GGobi features not only parallel coordinates but also scatter plots time series and bar charts The visualisations are synchronised record selections and data manipulations in one of them trigger an immediate update of the others There is also a table view of the input data set called data viewer which is however only partially synchronised with the visualisations User interaction in the parallel coordinates visualisation itself is rather limited In order to select records the user must move a fixed size rectangle over the the data points not the lines connecting
179. rsion 1 4 is used for unit tests Llopis and Nicholson 2011 Apart from the fact that Application has to link to all libraries there are a number of conceptual dependencies visualised in Figure 4 1 56 4 Multi Dimensional Explorer an ja mc General dependencies reste A Ld MDE s own module Se 3 party library Figure 4 1 Visualisation of the dependencies between MDE s modules and 3 party libraries An arrow indicates that a module needs to know about another module or 3 party library The Boost libraries are used throughout the entire application more as an extension of the C standard library than a separate module so dependencies on them are not documented explicitly Module Test contains tests of all other modules or is designated to contain tests of all other modules in future versions of MDE so conceptually it needs to know about all of the other modules and libraries Apart from the fact that these conceptional dependencies exist modules Application and Test are statically linked to the library files created from all other modules because they are compiled into executable files 4 2 Modules 57 4 2 Modules This section discusses each of the modules in MDE s software architecture in detail The order in which modules appear in this section is not related to their importance or level in the general design but is simply alphabetical For reasons of readability and in
180. ry limited compared to other parallel coordinates software as well Records are coloured from red to blue relative to their value in one of the dimensions which the user selects by clicking on the respective axis Dragging on an axis with the mouse creates an interval in which records are selected The data values of selected records are not shown anywhere 48 3 Parallel Coordinates 1613 lbs Figure 3 30 Parallel coordinates in Protovis Bostock and Heer 2009 a toolkit implemented in JavaScript and SVG This example shows Protovis running in a web browser and visualising the commonly used cars data set by Ramos and Donoh 1983 Performance is an issue with web based implementations of parallel coordinates and interaction methods are more limited than in other software 3 5 13 VizCraft VizCraft is an information visualisation tool written by Goel 1999 applying parallel coordinates to the domain of aircraft design but made such that any kind of multi dimensional data can be used with it It is implemented as a Java Oracle 2011 applet and can be used online from the official web page Figure 3 31 shows VizCraft visualising a small fictional test data set created by the tool s author Interaction is very limited in VizCraft Axes can be moved by dragging them with the mouse and they can be vertically zoomed In addition to that an axis can be chosen as the active one by clicking on it leading to records being coloured from
181. s 48 49 and 81 Google 2010 Google Body http bodybrowser googlelabs com Cited on page 3 Google 2011 Gadgets Google Docs Help http docs google com support bin topic py topic 15165 Cited on page 48 Graham Martin and Jessie Kennedy 2003 Using Curves to Enhance Parallel Coordinate Visual isations In Proc 9 IEEE Symposium on Information Visualization InfoVis 2003 pages 10 16 ISBN 0769519881 doi 10 1109 TV 2003 1217950 http www iidi napier ac uk c publications publicationid 2760350 Cited on pages 17 22 and 23 GTK Team 2010 The GTK Project http www gtk org Cited on page 42 Hackl Christian 2010 Parallel Coordinates Visualisation Term Paper at Graz University of Technol ogy Cited on pages 31 34 and 76 Halvey Martin J and Mark T Keane 2007 An Assessment of Tag Presentation Techniques In Proc 16 International World Wide Web Conference WWW2007 pages 1313 1314 ACM doi 10 1145 1242572 1242826 http www2007 org posters poster988 pdf Cited on page 3 Bibliography 123 Hauser Helwig Florian Ledermann and Helmut Doleisch 2002 Angular Brushing of Extended Par allel Coordinates In Proc 8 IEEE Symposium on Information Visualization InfoVis 2002 pages 127 130 IEEE Computer Society ISBN 076951751X ISSN 1522 404X doi 10 1109 INFVIS 2002 1173157 http www vrvis at publications pdfs PB VRVis 2002 035 paf Cited on pages 29 and 39 HCIL 2006 Un
182. s Tree giving access only to its const interface Modifying opera tions such as changeValue are intercepted so that registered Observer objects can be notified before the call is delegated to Tree This is the same design as in data Holder implementing the Observer design pattern by Gamma et al 1994 As for the technical structure of the tree a Node knows its children and its parent both via pointers A leaf node has an empty children collection while a root node stores the null pointer as its parent Table 4 5 lists all classes free standing functions and enums of module Options 4 2 5 Presentation This module contains components responsible for visual and textual presentation of the data structures in module Data Additionally it depends on the Presentation Structure module which it uses to create and maintain the geometrical structure of its visualisations and on the Options module to maintain user settings such as the colour of selected records 66 4 Multi Dimensional Explorer Name Purpose BatchChange Commits many option changes in one batch Change Changes made via Synchroniser ChildrenCountsDifferException Exception for differing children counts in checkEqual s node comparison Definition Defines an option DefinitionCountsDifferException Exception for differing definition counts in checkEqual s node comparison DefinitionNamesOrTypesDifferException Excepti
183. s a star hence the name the rays of each star corresponding to dimensions so that there are n rays for n dimensions The length of a ray depends on the value of the data point in the corresponding dimension relative to all other data points values in the same dimension The end points of a ray are connected to those of the two neighbouring rays Figure 2 8 shows the XmdvTool 2010 software displaying a star plot visualisation of the cereals data set provided by Velleman 1996 Fanea et al 2005 discuss an extension of star plots named parallel glyphs in which star plots are merged with parallel coordinates see Chapter 3 2 3 4 Visislands Andrews et al 2001 discuss an information visualisation technique they call VisIslands The goal of this technique is to reflect similarity between data vectors and display them such that a user can immediately spot similarities The dynamic creation of such a visualisation is an iterative process It involves a hierarchical clustering technique and clustering vectors in the data set around the centroids in terms of similarity The clustered vectors are then placed on a two dimensional rectangular area peaks displayed in red with circles in other colours around them The resulting graphics bear a certain resemblance to a map of islands in the sea this is where the name comes from Figure 2 9 shows a VisIslands visualisation of the data set about prices and earnings in different cities around t
184. s compensates for the lack of a feature to aggregate selections directly with the mouse such as by pressing the Shift or Ctrl key while clicking As a severe limitation the Picviz GUI makes it difficult to analyse the actual values of selected records There is a table displaying the whole data set but selections in the table and the parallel co ordinates visualisation are not synchronised In fact the table itself is completely inactive Values are only displayed as tooltips on the visualisation itself when the user places the mouse pointer on a record s polyline For any more complicated task the CLI frontend has to be used Finally anti aliasing can optionally be turned on to give users the choice to trade speed for display quality Additionally zooming is supported and the visualisation can be exported as a Portable Network Graphics PNG file 3 5 7 GGobi GGobi is a general purpose visualisation desktop application suitable for statistical analysis of mul tivariate data Cook and Swayne 2007 It is written in C using the GTK toolkit GTK Team 2010 and can read Extensible Markup Language XML and comma separated values CSV text files GGobi is open source and can be downloaded for free The application evolved from the older XGobi project Swayne et al 2007 Figure 3 25 shows the tool 3 5 Software Applications 43 GGobi Jelki File Display Interaction Tools Help SE x Region olive csv Parallel Coordinates cu
185. s controlling filtering are placed at the top and bottom of each axis These arrows can then be moved up and down restricting active that is selectable records to those whose lines run through the vertical area the interval between the arrows These active records are selected by clicking on their lines in the parallel coordinates visualisation Multiple records can be selected by pressing the Ctrl key while clicking Upon selection records appear in a different colour and the geographical and thematic similarity visualisations are updated accordingly Similarly the selection is synchronised with a standard tabular view of the data The other selection type found in InfoScope introduces an additional visual information channel in the graphical display by means of colour shading One dimension axis can be selected for this feature via the Color by dropdown menu upon which each record polyline is coloured according to its value in the selected dimension The colour shading scheme itself can be chosen too For example one can choose shading from black to red such that lower values tend more towards black while higher values tend more towards red There is no zooming available for the parallel coordinates visualisation its size always depends on the window size On the positive side anti aliasing can optionally be turned on Records are optionally focussed when hovering over their lines with the mouse pointer and simultaneously tooltips s
186. s selected Figure 3 15 illustrates the concept which Hauser et al 2002 call angular brushing In order to define such a query in Parallax the user has to click the A button in the tool bar then click on an axis line in the parallel coordinates visualisation two purple arrow handles appear on the axis line They can be dragged up and down to define the angle range A relative complement query is defined by the complement of two first layer queries If for instance the user chooses the relative complement between first layer queries A and B then the result of this second layer query are all records selected in A but not in B Second layer queries are combined logically with AND OR and NOT operators to form a first layer query Parallax offers an undo feature for some of its operations which represents an important plus in terms of usability On the downside the application s display quality suffers from the lack of anti aliasing The colour scheme can only be partially changed Furthermore a number of bugs detract from a flawless user experience Most importantly zooming results in seemingly random display errors to the point of rendering the application unstable and dialogue windows for choosing a query s associated colours will not always open The former error is acknowledged in the manual and may disappear in future versions of the application 1 0 at the time of writing 40 3 Parallel Coordinates 3 5 5
187. sects a rectan gle lessOrEqual Floating point aware less than or equal to comparison lineSegmentsIntersect Checks intersection of two line segments rgbToHsl Converts from RGB to HSL svgCircle SVG code for a circle svgClipPath SVG code for clipping svgColour SVG code for an RGB colour svgEllipse SVG code for an ellipse svgLine SVG code for a line svgPolygon SVG code for a polygon svgRectangle SVG code for a rectangle svgText SVG code for text svgTriangle SVG code for a triangle Table 4 8 The classes and free standing functions of MDE s Presentation Structure module 76 4 Multi Dimensional Explorer cities2009 csv Multi Dimensional Explorer EJES ioj xi Amsterdamt16 8 1105 1125 1000 1255 1198 2373 300 158 84 66 477 1130 14455 18703 57100 76700 72300 112200 112000 93600 93700 58300 43000 48500 50200 TWN Zurich 120 145 309 249 50 61 1582 60 12 8 8 9 do 100 sido 4300 6200 ss 4700 2500 200 1800 1200 180 Pr la 4 Sy S amp Fa ay E a TM Song er de o ona a a a O o be o A o Meng ee nat ER a rg on enag one 4 ha Ory Y 78 ney m Me ey Vln a tag Slag o h go 4 9 0 a Me So en raring ag bats eo K tan u Po Moray rg Orr Pry er gy Mo Rn Mea erty Per Uso vw Ore te m oon Sig o Wr Poy so mi Us Uso Yea so Yow a Von KA en Ore ee mo e Do ms Sey KA m Yea
188. sentation Structure module is Visualisation Structure a collection of VisualisationElement instances with related operations for iteration and specific querying VisualisationElement has a memory efficient internal implementation in the form of a discrimi nated union Just as in the case of data Point the use of a union here is justified by the sheer number of objects which can be instantiated from this class There may literally be millions of visualisation Element objects all created while the visualisation is also being rendered to the screen For reasons of safety in face of union usage however VisualisationStructure alone is allowed to create new instances of VisualisationElement Other parts of MDE can instantiate the class only indirectly via public methods of visualisationStructure such as addLine Or addCircle The contents of VisualisationStructure are inspected for 1 Geometrical operations with the current mouse pointer position in InteractiveVisualisation such as finding out whether a particular VisualisationElement is under the mouse pointer 2 Exporting a visualisation as an SVG file Figure 4 9 shows an example of what could be contained in a VisualisationStructure and how buildsvg would transform the contents into SVG code Table 4 8 lists all classes and functions of the Presentation Structure module Functions rgbToHs1 and hs1ToRgb used for making an RGB colour lighter are based on JavaSc
189. sions In the latter case the value is the index of the enumeration value Matrix also stores information about entire dimensions in a separate collection Of Dimension instances Almost all of MDE operates only on Matrix conste using methods like pointString and pointNormalised to obtain a string representation of a field or its normalisation in the range 0 0 1 0 irrespective of its dimension s type Modifying operations are mirrored in the inter face of class Holder which delegates them to Matrix and reports the change to any registered Observer 4 2 Modules ain Er Figure 4 3 Every field stored in a Matrix is presented to the outside world in a unified manner either as a string representation or as a normalised value in the range 0 0 1 0 This is a safe interface to a more complex internal structure Every instance Of Dimension is either of type Integer Real or Enumeration and according to that type different member variables both in Dimension and in corresponding Field objects are either active or inactive In this example dimension A is of type Integer and therefore stores maximum and minimum integer values to define its range Those are used for normalising individual values 30 here is normalised to 0 3 B works similarly for real numbers C is of type Enumeration and therefore stores a vector of possible values an index into that vector stored in individual fields In this example
190. some its features The following list discusses major differences with other tools in detail e MDE takes its user interface approach from MDVS Hack 2010 It provides a Multiple Docu ment Interface MDI Microsoft 2010a in which different visualisations reside in different sub windows Textual overviews of individual records and individual dimensions are integrated con sistently into this scheme Record selection is achieved by drawing a selection rectangle with the 4 3 Comparison With Other Tools 77 Name Purpose TestChange Tests isolated data Change TestChangeDimensionType Tests changing dimension types in data Matrix es tChangeMat rixReplacement Tests data Change for replacement of matrix in a data Holder es t ChangeMinMax Tests changing min max values in data Matrix es tChangeRecordState Tests changing record state in data Matrix es tCorrectPointStringConverter Tests data PointStringConverter with correct decimal point characters TestDefaultDimensionFilters Tests default instances of data DimensionFilter for data Holder TestDimension Tests isolated data Dimension TestDimensionFilter Tests isolated data DimensionFilter TestDimensionIndexAndPosition Tests data DimensionIndex and data DimensionPosition TestDimensionReordering Tests changing the dimension order in data Matrix TestDimensionRe
191. splay lists and uses basic pixel operations for its caching mechanism rather than relying on texture operations Both of those techniques promise further performance gains Additionally MDE s caching mechanism currently supports only a fixed number of cache levels the mechanism could instead be generalised to support n cache levels which would cer tainly prove interesting for interactive parallel coordinates visualisations where operations often update only a horizontal subrange of the screen when axes are being moved Another use for a dynamic number of cache levels is to put polylines for unused records into an intermediate layer between the background and the main contents The current design always considers unused records part of the main contents This keeps the user interface responsive when filter handles are being dragged with the mouse because the background does not have to be recreated but sacrifices performance gains which would be obtained otherwise for the case when a user has finished setting his or her filters and continues modifying only selections because in this case the current implementation of MDE redraws the same unused record lines over and over An intermediate layer would combine responsiveness with additional performance gains at the cost of slightly increased memory consumption MDE does not currently feature clustering either This is partly due to the fact that little research has gone so far into the translati
192. sses commonly used procedures in exploratory analysis with interactive parallel co ordinates The discussion is based on the tutorial of Bauer et al 2010 created for Keith Andrews course on information visualisation at Graz University of Technology Andrews 2010 However the tutorial contained in this thesis uses screenshots from the Multi Dimensional Explorer MDE discussed in Chapter 4 and InfoScope Brodbeck and Girardin 2003 discussed in Section 3 5 2 Furthermore it uses a different data set for its examples Bauer et al use data about the correlation between smoking and cancer Fraumeni 1968 whereas this thesis uses data about prices and earnings in different cities provided by UBS 2009 The data set was turned into a comma separated value file suitable for opening in MDE and many other software applications with InfoScope Table 3 1 shows part of the cities data set 3 4 1 Removing Axes Often not all the dimensions of a multi dimensional data set are of interest to the task at hand Therefore it makes sense to view only the axes for important dimensions and remove the others Figure 3 10 shows the entire data set with all 52 dimension axes Figure 3 11 shows the parallel coordinates visualisation after removal of those axes which the user is not interested in All but 5 of the 52 dimensions in the cities data set have been hidden by the user leaving city names paid vacation days per year time required for 1 kg of brea
193. sualisation technique this thesis focuses on 2 1 History Information visualisation might appear a field originating in computer science but in truth dates back to earlier centuries and is a concept much older than computer science itself In the second half of the 18 and the first half of the 19 century British statistician William Playfair published diagrams based on statistical data about England s trade relations Playfair 2005 According to FitzPatrick 1960 he is considered one of the pioneers of information visualisation German Original Wenn ich Sie auffordere Bitte beobachten Sie so sollten Sie mich dem Sprachgebrauch gem fragen Ja aber was Was soll ich beobachten Mit anderen Worten Sie bitten mich Ihnen ein Problem anzugeben das durch Ihre Beobachtung gel st werden kann English translation by the author of this thesis 4 2 Information Visualisation DIAGRAM or rus CAUSES or MORTALITY APRIL 1855 ro MARCH 1856 IN THE ARMY n THE EAST APRIL 1854 ro MARCH 1855 Jury o wow Figure 2 1 Florence Nightingale s wedge diagram from 1857 is a historical example of infor mation visualisation long before computers came into existence The visualisation shows the majority of soldiers in the Crimean Wars died because of poor sanitary conditions rather than being killed by the enemy in combat Image copyright of the Wellcome Library 201 1a The wedge diagram by famous British
194. tains no new definitions or overridden values while My View contains a special option dot_size and overrides background_colour This leads to the following options being available and values being set for them 4 2 Modules 65 background colour FFFFFF label_font Verdana overridden background colour 000000 dot_size 15 new definition Figure 4 5 MDE s tree based options mechanism supports inheritance and overriding In this example Root My View and Your View are Node objects Root being the parent of the other two Root contains two options background_colour and label font which the others inherit My View overrides the parent value of background_colour and adds dot_size an option definition of its own e Root background_colour FFFFFF label_font Verdana e My View background_colour 000000 label_font Verdana dot_size 15 e Your View background_colour FFFFFF label_font Verdana As this example shows which definitions are available and which values they are currently set to depends on which node is being queried In common Design Patterns terminology Gamma et al 1994 Tree acts as a Facade for the more complex Node class Clients of Tree are not completely oblivious of Node but they only have to deal with const references not with pointers or non const references if they do not explicitly request them Class synchroniser encapsulate
195. th wrong filename estCsvFileReaderWrongParameters Tests data input CsvFileReader with wrong parameters estFileReaderGroup Tests data input FileReaderGroup TestIsDefaultValidLine Basic tests for data input IsDefaultValidLine estMakeDimensionSimpleColumns Tests data input makeDimension O with simple input estMatrixCreationFromStream Tests creation of data Matrix from stream TestMatrixCreationRowTooLong Tests creation Of data Matrix with input row being too long estMatrixCreationRowTooShort Tests creation of data Matrix with input row being too short TestTextStreamInput Basic tests for data input TextStreamInput TestTextStreamInputComplex Complex tests for data input TextStreamInput TestTextStreamInputEmpt y Tests empty string input for data input TextStreamInput TestTextStreamInputOnlyInvalidLines Tests invalid lines input for data input TextStreamInput Table 4 10 The unit tests in MDE s Test module testing components of module Data Input 4 3 Comparison With Other Tools 79 Name Purpose TestColour Basic tests for presentation structure Colour TestOptionDefinition Basic tests for options Definition TestOptionNode Tests for isolated options Node TestOptionsChange Tests isolated options Change TestReadXml Tests options r
196. tion visualisation is to provide insights into data sets using graphical representations of the data which would be difficult to extract otherwise Lex 2008 notes that a distinction must be made between information visualisation which deals with abstract information and scientific visualisa tion which deals with information bearing intrinsic visual properties Google Body Google 2010 is an example of a visualisation tool not meeting the criteria of information visualisation because the three dimensional graphics it produces do not visualise abstract information but are instead just a rendering of the real structure of the human body A better example of information visualisation would be tag clouds as discussed by Halvey and Keane 2007 because the graphical attributes of that visualisation scheme such as position colour and size of words are derived from non visual abstract information about the relationship between elements of the cloud Modern computer applications excel in the field of information visualisation due to their sheer cal culation power in creation of graphics Nevertheless information visualisation existed long before com puters were invented as is explained in Section 2 1 Section 2 2 talks about interactivity playing a major role in modern age information visualisation Section 2 3 discusses some popular techniques for multi dimensional data visualisation Chapter 3 then discusses parallel coordinates the information vi
197. tions in the field Chapter 4 discusses the original software developed for this thesis based on the experience gained from using the other software tools 3 1 Background Parallel coordinates did not originate in the field of information visualisation They were invented as a purely mathematical concept the transformation of orthogonal axes into parallel ones and can be traced back to 1885 when Maurice d Ocagne a French mathematician published his book Coordonn es parall les et axiales M thode de transformation g om trique et proc d nouveau de calcul graphique d duits de la consid ration des coordonn es parallelles Parallel and axial coordinates A geometric transformation method and a new graphical calculation procedure derived from the consideration of parallel coordinates d Ocagne 1885 d Ocagne applied parallel coordinates to numerical problems It was not until 1959 that Alfred Inselberg rediscovered them independently and applied them to information visualisation Inselberg 2010b Quoting Inselberg himself in Inselberg 2007 p 645 I wondered why geometry was being tackled without using the fun and benefits of pictures What is sacred about orthogonal axes which quickly use up the plane associated with them Inselberg s publications Inselberg 1985 Inselberg and Dimsdale 1990 would later help spread the concept of parallel coordinates in the field of information visu
198. tionsDialoguePanel Page in an OptionsDialogue RecentFiles Container storing paths of recently opened files RecentFilesIterator Iterator for RecentFiles MainWindowIdentifier Identifiers for tool bar and menu items in MainWindow createDefaultFileReaderGroup Creates default data input FileReaderGroup createOptionsFileError Creates options file error message handleException Displays current exception s error message itemIdentifierDataViewHighest Highest of all add data view menu item identi fiers itemIdentifierDataViewLowest Lowest of all add data view menu item identifiers openFile Opens file with data for a data Matrix serialiseMdiChildFrames Generates string from DataViewMdiChildFrame instances position and size Table 4 2 The classes enums and free standing functions of MDE s Application module 4 2 Modules 59 4 2 2 Data This module contains data structure components and means to manipulate them independent of their visual or textual representation It also features mechanisms to intercept modifications of the data and notify other parts of the software of any such changes Data is entirely self contained and does not know about or depend on any of the other modules The way MDE internally represents multi dimensional data and how it is stored and synchronised is explained in Figure 4
199. tiveVisualisation stems mainly from the desire to avoid a central class from becoming too huge Their purposes are not entirely separated from each other the result being mutual friendship Presentation is clearly the most complex of MDE s modules Two particularly complex areas are pixel caching and rendering of fonts in OpenGL These are discussed in detail in Chapter 5 See also Appendix A for more detailed information on how to add new visualisation subclasses to MDE Table 4 6 is a list of the classes structs and free standing functions of the Presentation module which are not related to any of the concrete Dataview classes Table 4 7 on the other hand lists all components of the module related specifically to the parallel coordinates visualisation the scatter plot visualisation and the textual record and dimension views 4 2 6 Presentation Structure Presentation Structure a sub module of Presentation contains the components responsible for the logical structuring of data presentations This includes Scalable Vector Graphics SVG support Eisenberg 2002 Ramani et al 2008 Presentation Structure is entirely self contained and does not depend on any other MDE module The structure of a presentation is different from the structure of data A presentation s structure is defined in geometrical terms such as the end points of lines or the centre and radius of circles and related geometrical operations The central class of the Pre
200. to add new visualisations to MDE using a concrete but simple example The audience of this developer guide are not end users of MDE End users should instead refer to the User Guide The example of a new imaginary visualisation discussed in this guide is as follows for each record a horizontal line across the full window width is displayed its vertical position depending on the first di mension s data value its colour depending on whether the corresponding record is used black unused grey or selected blue Figure B 1 shows the desired result Note that this is not a real information visualisation technique or at least not a very useful one the example was chosen for its simplicity so that the developer guide can concentrate on programming aspects B 1 Development Environment Setup MDE s source code is managed by the Subversion SVN version control software Apache Software Foundation 2011 After a fresh checkout of the repository the following subdirectories appear application e data e data input options e presentation presentation structure test Italian Original Quello che scrivete e come lo scrivete pud cambiare la vostra vita English translation by the author of this thesis 109 110 B MDE Developer Guide Pre AE SVG Export zoom 100 Options Figure B 1 An example of a new visualisation added to MDE For each record of the data set a horiz
201. ual modules of the MDE source code Section 4 3 compares MDE to other parallel coordinates software discussed previously Chapter 5 then discusses two selected areas in MDE s software design which are particularly com plex in the overall architecture Appendix A is a manual for MDE end users and explains all of its features in detail Appendix B gives future developers detailed information on how to extend MDE with new visualisations 4 1 Software Architecture MDE is written in C conforming to the 2003 ISO standard of the language International Organization for Standardization 2003 No C 0x features International Organization for Standardization 2010 are used Such features are already available in many C compilers and might keep some parts of the source code more concise and could in fact be employed in future versions of MDE but the new language standard is still too much of a work in progress and its implementations are still too experimental to meet MDE s goal of being a robust and reliable piece of software The C code of MDE follows generally accepted principles of good software engineering in C as defined by Meyers 1996 1997 using standard design patterns by Gamma et al 1994 The major principles in MDE s software design and implementation are 53 54 4 Multi Dimensional Explorer 1 Use of the C standard library whenever possible Josuttis 1999 For example no use of a custom string class instead of
202. ues of unknown options in options file Type Type of an option checkEqual Checks two Node objects and their descendents for an equal structure getAvailableFonts Retrieves fonts available on the system getDefaultFontName Retrieves default font name getFontDirectory Retrieves the system directory for font files getFontPath Retrieves the complete path of a font file readOptionsFile Reads the options file readXml Reads options XML code writeXml Writes options XML code Table 4 5 The classes free standing functions and enums of MDE s Options module 4 2 Modules 67 MDE can be extended with additional visualisations by subclassing the abstract visualisation base class which is itself a specialisation of the more general abstract Dataview The abstract class Data View is implemented for non visual ways of data presentation in the sense of no graphical rendering InteractiveVisualisation is a predefined derivee of Visualisation providing interaction with the mouse and keyboard for the user of a visualisation window DataView is derived from the wxPane1 class of the wx Widgets framework but multiply inherits also data Observer and options Observer so that all views will always be notified when data changes occur in one of them and when program options have been modified by the user Multiple inheritance is used in a safe manner here due to data Observer and optio
203. ule contains two implementations Of VisualisationImplementation VisualisationImplementationOpenGL is MDE s main rendering engine It is implemented in terms of OpenGL Shreiner 2009 Khronos Group 201 La e VisualisationImplementat ionGraphicsRenderer is an alternative engine which uses the oper ating system s standard Application Programming Interface API for graphics drawing This class should be considered more as a proof of concept for MDE s software architecture supporting dif ferent rendering backends It is much slower than the OpenGL backend The relationship between Visualisation and VisualisationImplementation is such that the for mer instantiates an instance of the latter visualisation is the graphical frontend visualisation Implementation is the backend This is a classical application of the Bridge design pattern by Gamma et al 1994 System specific and low level graphics operations are the concern of Visualisation Implementation while visualisation is concerned with providing a protected interface with which derived classes can build concrete visualisation schemes 4 2 Modules 69 application MainStatusBar onNewData V data options wxPanel Observer Observer onNewData onNewOption A presentation presentation DataView prese
204. user action in parallel coordinates if the change of a record s state triggered a rebuild of the entire visualisation then the entire visualisation would have to be rebuilt every time a filter handle was used resulting in a slow and unresponsive user interface 85 86 5 Selected Details of the Implementation As long as there are still records to be processed createVisualisation is called This function first checks the number of records which have been processed already if it is O and if no back ground cache exists performed by inspecting the boolean background_cached_ variable which is initially false then further control is delegated to the virtual function visualisationFirst Step which concrete subclasses implement to render the background of the visualisation Sub classes can also clear the background cache by calling clearBackgroundCache if they find that data or option changes have invalidated it When ParallelCoordinatesVisualisation has finished creating the background visualisation ImplementationOpenGL caches the current pixel output by saving it in its background_ member variable Visualisation Sets background_cached_to true If createVisualisation is called with an active background cache the renderBackgrouna method of VisualisationImplementationOpenGL is called first pasting the cached background pixels on the window followed by a call to visualisationFirstStep with any actual rendering disabled
205. versions may be used and should work ifthey are guaranteed to be backwards compatible It is necessary that all libraries mentioned above are built from source code The reason for this re quirement is that MDE itself is built with the non default compiler switch mr Multi Threaded Runtime Library Therefore any libraries used must have been built with the same compiler switch Failure to do so will result in linker errors upon any attempt to build MDE SVN ignore lists have been set for all files and directories the build process might generate such as various Release and Debug subdirectories binaries or user files No attempts to commit any of those should be made B 2 Preparing a New Visualisation 111 zox File Edit View Project Build Debug Tools Window Help regal aid e ri Release win32 I s a meena ii QA External Dependencies ij Headerdateien A Quelldateien deal mde presentation structure a mde test Show output from Buld Jalea lala 7 gt visualisation_structure_iterator cpp 7 gt visualisation_structure_records cpp wrong_point_string_converter cpp Generating code Finished generating code test vcxproj gt D christian mde Release mde test exe Success 52 tests passed Test time 22 seconds Build 7 succeeded failed up to date skipped a Solution Explorer Ka Se EE377 m Property Manager ER Code Definition Window ER AAA Output SEAS Build suc
206. was outlined usability is becoming an ever greater concern in this area With all this context established Chapter 3 explained parallel coordinates Ever since Alfred Insel berg reintroduced the technique in the second half of the 20 century it has become ever more popular and more widely used Parallel coordinates are a unique information visualisation technique because they impose no theoretical limit on the number of dimensions which can be displayed at the same time The thesis explained why parallel coordinates work like this and why they differ from any orthogonal approach However the thesis also outlined problems with parallel coordinates The most important problem is that large amounts of data result in the visualisation being cluttered with lines As the thesis demonstrates many techniques to mitigate this problem have been invented by the scientific community the solution proposed is usually to combine records or lines into clusters Random sampling using neural networks or using time animation have all been suggested as extensions to the basic parallel coordinates technique This thesis explained also how to use parallel coordinates It first discussed basic interaction features and then reviewed a large number of software applications featuring parallel coordinates to demonstrate that the basic concept of the technique has been implemented in many different ways The thesis reviewed each of the software applications strengths
207. way of synchro nising data in different visualisation views is similar though MDE on the other hand provides feature rich user interaction not found in GGobi such as simple record selection and aggregation of selections or zooming of the display Another major difference is MDE s support for anti aliasing XmdvTool XmdvTool 2010 features clustering and time series animation two features not found in MDE It uses OpenGL as its rendering backend but in contrast to MDE disables anti aliasing MDE makes it easier to select and identify individual records as visualisations can be viewed next to each other and are also synchronised with textual record and dimension tables absent from XmdvTool VisDB by Keim et al 2002 offers a set of different synchronised visualisations just like MDE but it provides little support for user interaction MDE features many ways of selecting records defining filters moving and flipping dimensions and zooming all of this is absent from VisDB 4 3 Comparison With Other Tools 81 e Caleydo by Lex et al 2010 is rooted in medical science whereas MDE is general purpose soft ware which makes it simpler to use for data analysis in other domains without distraction by features useful only for Caleydo s original purpose Caleydo in contrast to MDE features angular brushing but does not have a simple mouse selection rectangle mechanism like MDE OECD eXplorer Organisation for Economic Co operation an
208. xt contents of the input file the table on the right part is a preview of how it will be interpreted A 2 Parallel Coordinates View 97 A 2 Parallel Coordinates View The parallel coordinates view visualises multi dimensional input data with the parallel coordinates vi sualisation technique which involves arranging axes in parallel to each other on a plane each axis rep resenting one of the dimensions Records are represented by polylines connecting the axes Figure A 2 shows MDE implementing the concept A number of general display options for parallel coordinates can be changed in MDE using the controls in the view s subwindow From left to right An SVG Export button with which the visualisation is saved as an SVG file A Zoom spin control for selecting the visualisation s zoom level Controls to change the opacity of the polylines 50 by default Controls to change the angle by which axis names are rotated or to turn rotation off completely and instead truncate names so that they do not overlap Controls to individually turn on and off polylines the mean polyline and axis histograms An Options button to fine tune further options such as the colours of individual visualisation ele ments Figure A 3 shows these controls Controls for SVG export zooming and the Options button make sense for all graphical views and therefore appear also in the scatter plot visualisation The first thing a user will want to do with a

Multi-Dimensional Explorer - Das IICM

Contents

Download Pdf Manuals

Related Search

Related Contents