1Parallel Coordinates ++CS 4460/7450 - Information VisualizationFeb. 2, 2010John StaskoLast Time• Viewed a number of techniques for portraying low-dimensional data (about 3<x<20) scatterplot matrix Table Lens sliding rods Attribute Explorer Dust & Magnet etc.Spring 2010 CS 4460/7450 22Spring 2010 CS 4460/7450 3Parallel Coordinates• What are they? Explain…Spring 2010 CS 4460/7450 4Parallel CoordinatesV1 V2 V3 V4 V52 7 6 3 49 8 1 4 27 3 4 8 1D1D2D33Spring 2010 CS 4460/7450 5Parallel CoordinatesV1 V2 V3 V4 V57 3 4 8 1D1012345678910Spring 2010 CS 4460/7450 6Parallel CoordinatesV1 V2 V3 V4 V5D22 7 6 3 40123456789104Spring 2010 CS 4460/7450 7Parallel CoordinatesV1 V2 V3 V4 V5D39 8 1 4 2012345678910Spring 2010 CS 4460/7450 8Parallel CoordinatesV1 V2 V3 V4 V5Encode variables alonga horizontal rowVertical line specifies differentvalues that variable can takeData point represented as apolyline5Spring 2010 CS 4460/7450 9Parallel Coords ExampleBasicGrayscaleColorSpring 2010 CS 4460/7450 10Issue• Different variables can have values taking on quite different ranges• Must normalize all down (e.g., 0->1)6Spring 2010 CS 4460/7450 11Application• System that uses parallel coordinates for information analysis and discovery• Interactive tool Can focus on certain data items ColorTaken from:A. Inselberg, “Multidimensional Detective”InfoVis „97, 1997.Spring 2010 CS 4460/7450 12Discuss• What was their domain?• What was their problem?• What were their data sets?7Spring 2010 CS 4460/7450 13The Problem• VLSI chip manufacture• Want high quality chips (high speed) and a high yield batch (% of useful chips)• Able to track defects• Hypothesis: No defects gives desired chip types• 473 batches of dataSpring 2010 CS 4460/7450 14The Data• 16 variables X1 - yield X2 - quality X3-X12 - # defects (inverted) X13-X16 - physical parameters8Spring 2010 CS 4460/7450 15Parallel Coordinate DisplayYikes!But notthat badyield &qualitydefectsparametersDistributionsx1 - normalx2 - bipolarSpring 2010 CS 4460/7450 16Top Yield & QualitydefectsHave some defectssplit9Spring 2010 CS 4460/7450 17Minimal DefectsNot thehighestyields andqualitySpring 2010 CS 4460/7450 18Best YieldsAppears thatsome defectsare necessaryto produce the best chipsNon-intuitive!10Spring 2010 CS 4460/7450 19XmdvToolToolsuite createdby Matthew Wardof WPIIncludes parallelcoordinate viewsParVis SystemSpring 2010 CS 4460/7450 20http://www.mediavirus.org/parvis/Demo11Spring 2010 CS 4460/7450 21ChallengesOut5d dataset (5 dimensions, 16384 data items)(courtesy of J. Yang)Too muchdataSpring 2010 CS 4460/7450 22Dimensional ReorderingSame dimensions ordered according to similarityYang et alInfoVis ‟03Which dimensions are most like each other?12Spring 2010 CS 4460/7450 23Dimensional ReorderingPeng et alInfoVis „04Can you reduce clutter and highlightother interestingfeatures in data bychanging order ofdimensions?Spring 2010 CS 4460/7450 24Reducing DensityJerding and Stasko, ‟95, ‟98Wegman & Luo, ‟96Artero et al, 04Johansson et al, „0513Improved Interaction• How do we let the user select items of interest?• Obvious notion of clicking on one of the polylines, but how about something more than thatSpring 2010 CS 4460/7450 25Spring 2010 CS 4460/7450 26Attribute Ratios• Angular Brushing Select subsets which exhibit a correlation along 2 axes by specifying angle of interestHauser, Ledermann, & DoleischInfoVis „02(earlier demo)14Spring 2010 CS 4460/7450 27Range Focus• Smooth Brushing Specify a region of interest along one axisSpring 2010 CS 4460/7450 28Combining• Composite Brushing Combine brushes and DOI functions using logical operators15Spring 2010 CS 4460/7450 29Videohttp://www.vrvis.at/via/research/ang-brush/parvis4.movApplicationSpring 2010 CS 4460/7450 30http://www.syracuse.com/news/index.ssf/2010/01/data_mining_helps_new_york_cat.html16Different Kinds of Data• How about categorical data? Can parallel coordinates handle that well?Spring 2010 CS 4460/7450 31Spring 2010 CS 4460/7450 32Parallel Sets• Visualization method adopting parallel coordinates layout but uses frequency-based representation• Visual metaphor Layout similar to parallel coordinates Continuous axes replaced with boxes• Interaction User-driven: User can create new classificationsKosara, Bendix, & HauserTVCG „0517RepresentationSpring 2010 CS 4460/7450 33Color used for differentcategoriesThose values flow into theother variablesSpring 2010 CS 4460/7450 34ExampleTitanic passengersdata set18Spring 2010 CS 4460/7450 35Titanic Data SetInteractionsSpring 2010 CS 4460/7450 3619Spring 2010 CS 4460/7450 37VideoInfoVis „05Spring 2010 CS 4460/7450 38Star PlotsVar 1Var 2Var 3Var 4Var 5ValueSpace out the nvariables at equalangles around a circleEach “spoke” encodesa variable‟s valueAlternative Rep.Data point is now a “shape”20Spring 2010 CS 4460/7450 39Star Plot exampleshttp://seamonkey.ed.asu.edu/~behrens/asu/reports/compre/comp1.htmlSpring 2010 CS 4460/7450 40Star Coordinates• Same ideas as star plot• Rather than represent point as polyline, just accumulate values along a vector parallel to particular axis• Data case then becomes a point21Spring 2010 CS 4460/7450 41Star CoordinatesE. Kandogan, “Star Coordinates: A Multi-dimensional VisualizationTechnique with Uniform Treatment of Dimensions”, InfoVis 2000Late-Breaking Hot Topics, Oct. 2000DemoSpring 2010 CS 4460/7450 42Star Coordinates• Data cases with similar values will lead to clusters of points• (What‟s the problem though?)• Multi-dimensional scaling or projection down to 2D• Good segue to next time…22Spring 2010 CS 4460/7450 43Parallel Coordinates• Technique Strengths? Weaknesses?Spring 2010 CS 4460/7450 44Design Project• For graduate students• Group of 2-4 students Can argue for singleton• First milestone: Teams and topics in 2 weeks• Choose from list of topics Can argue for your own23Spring 2010 CS 4460/7450 45Upcoming• High-dimensional data representations Reading:Yang et al, InfoVis ‟04• InfoVis Systems & Toolkits ReadingViegas et al, TVCG