1 Multivariate Visual Representations 2 CS 7450 - Information Visualization Sep. 16, 2013 John Stasko Topic Notes Fall 2013 CS 7450 2 Recap • We examined a number of techniques for projecting >2 variables (modest number of dimensions) down onto the 2D plane  Scatterplot matrix  Table lens  Parallel coordinates  etc.2 Fall 2013 CS 7450 3 Varieties of Techniques Another Type of Data • Temporal, with different types/categories taking on values at the various points in time Fall 2013 CS 7450 43 Baby Names • We saw a demo back at the start of the term • M. Wattenberg developed a visualization to help promote his wife’s book on the topic • Used 100+ years of US Census data on baby names • Became an internet rage  500,000 hits in first two weeks Fall 2013 CS 7450 5 Wattenberg & Kriss TVCG ‘06 Fall 2013 CS 7450 6 The Visualization • Shneiderman’s mantra • Dynamic Query Approach • Keyboard-based mechanism for filtering • Pop-up boxes for details • Smooth animation on each transition http://babynamewizard.com/namevoyager/ Stacked bargraph  StreamGraph4 Fall 2013 CS 7450 7 Examples Result of typing O Result of typing Unkown Dust & Magnet • Altogether different metaphor • Data cases represented as small bits of iron dust • Different attributes given physical manifestation as magnets • Interact with objects to explore data Fall 2013 CS 7450 8 Yi, Melton, Stasko & Jacko Information Visualization ‘055 Interface Fall 2013 CS 7450 9 Interaction • Iron bits (data) are drawn toward magents (attributes) proportional to that data element’s value in that attribute  Higher values attracted more strongly • All magnets present on display affect position of all dust • Individual power of magnets can be changed • Dust’s color and size can connected to attributes as well Fall 2013 CS 7450 106 Interaction • Moving a magnet makes all the dust move  Also command for shaking dust • Different strategies for how to position magnets in order to explore the data Fall 2013 CS 7450 11 See It Live Fall 2013 CS 7450 12 Video & Demo ftp://ftp.cc.gatech.edu/pub/people/stasko/movies/dnm.mov7 Set Operations • Different type of problem  Large set of items, each can be in one or more sets  How do we visually represent the set membership? Fall 2013 CS 7450 13 Standard Technique Fall 2013 CS 7450 14 Venn Diagram A B C Contains all possible zones of overlap8 Alternately Fall 2013 CS 7450 15 Euler Diagram Does not necessarily show all possible overlap zones http://en.wikipedia.org/wiki/File:British_Isles_Euler_diagram_15.svg But what’s the problem? Bubble Sets Fall 2013 CS 7450 16 Collins et al TVCG (InfoVis) ‘09 Video9 ComED & DupED Fall 2013 CS 7450 17 Riche & Dwyer TVCG (InfoVis) ‘10 Video Fall 2013 CS 7450 18 Step Back • Most of the techniques we’ve examined work for a modest number of data cases or variables  What happens when you have lots and lots of data cases and/or variables?10 Fall 2013 CS 7450 19 Many Cases Out5d dataset (5 dimensions, 16384 data items) (courtesy of J. Yang) Many Variables Fall 2013 CS 7450 2011 Strategies • How are we going to deal with such big datasets with so many variables per case? • Ideas? Fall 2013 CS 7450 21 General Notion • Data that is similar in most dimensions ought to be drawn together  Cluster at high dimensions • Need to project the data down into the plane and give it some ultra-simplified representation • Or perhaps only look at certain aspects of the data at any one time Fall 2013 CS 7450 2212 Fall 2013 CS 7450 23 Mathematical Assistance 1 • There exist many techniques for clustering high-dimensional data with respect to all those dimensions  Affinity propagation  k-means  Expectation maximization  Hierarchical clustering Fall 2013 CS 7450 24 Mathematical Assistance 2 • There exist many techniques for projecting n-dimensions down to 2-D (dimensionality reduction)  Multi-dimensional scaling (MDS)  Principal component analysis  Linear discriminant analysis  Factor analysis Data mining Knowledge discovery Comput Sci & Eng courses Visual Analytics, Prof. Lebanon13 Fall 2013 CS 7450 25 Other Techniques • Other techniques exist to manage scale  Sampling – We only include every so many data cases or variables  Aggregation – We combine many data cases or variables  Interaction (later) • Employ user interaction rather than special renderings to help manage scale Fall 2013 CS 7450 26 Our Focus • Visual techniques • Many are simply graphic transformations from N-D down to 2-D14 Fall 2013 CS 7450 27 Use? • What kinds of questions/tasks would you want such techniques to address?  Clusters of similar data cases  Useless dimensions  Dimensions similar to each other  Outlier data cases  … • Think back to our “cognitive tasks” discussion Fall 2013 CS 7450 28 Now • We’ll examine a number of other visual techniques intended for larger, higher-dimensional data sets15 Fall 2013 CS 7450 29 Can We Make a Taxonomy? • D. Keim proposes a taxonomy of techniques  Standard 2D/3D display Bar charts, scatterplots  Geometrically transformed display Parallel coordinates  Iconic display Needle icons, Chernoff faces  Dense pixel display What we’re about to see…  Stacked display Treemaps, dimensional stacking TVCG ‘02 Minimum Possible? • We have data cases with variables • What’s the smallest representation we can use?  How? Fall 2013 CS 7450 3016 Fall 2013 CS 7450 31 Dense Pixel Display • Represent data case or a variable as a pixel • Million or more per display • Seems to rely on use of color • Can pack lots in • Challenge: What’s the layout? Fall 2013 CS 7450 32 One Representation • Grouping arrangement • One pixel per variable • Each data case has its own small rectangular icon • Plot out variables for data point in that icon using a grid or spiral layout Uses color scale17 Fall 2013 CS 7450 33 Illustration Levkowitz Vis ‘91 Related Idea • Pixel Bar Chart • Overload typical bar chart with more information about individual elements Fall 2013 CS 7450 34 Keim et al Information Visualization ‘0218 Fall 2013 CS 7450 35 Idea 1 Height encodes quantity Width encodes quantity Idea 2 • Make each pixel within a bar correspond to a data point in that group represented by the bar  Can do