Docstoc

Visualization

Document Sample
Visualization Powered By Docstoc
					Information Visualization –III
 Treemaps and Fisheye Views
          Yaji Sripada
   In this lecture you learn
• Visualizing large amounts of
  information in small display area
• Visualizing large amounts of
  hierarchical information
  – TreeMaps
• A general strategy to Visualizing
  large amounts of information
  – Fish eye views

   Dept. of Computing Science, University of Aberdeen   2
                      Introduction
• Common challenge in designing modern
  infovis tools is
  – To visualize large quantities of information in
    small display area
• Two popular solutions
  – Treemaps (not Java TreeMaps)
      • Visualizing large amounts of hierarchical information
  – Fisheye views
      • Visualizing large amounts of any type of information
        with known user degree of interest (DOI)


    Dept. of Computing Science, University of Aberdeen          3
            Visualizing Hierarchical
                  Information
• A lot of information in the real world is structured
  hierarchically
   –   File system structure on an OS such as UNIX
   –   Family Trees
   –   User Manuals
   –   Computer programs
   –   Etc
• Hierarchical information structure is made up of
   – Links and
   – Nodes
• Common solutions for visualizing hierarchical
  information
   – Listings e.g directory listings on UNIX
   – Outlines e.g document outline in MSWord
   – Tree diagrams e.g windows explorer

       Dept. of Computing Science, University of Aberdeen   4
         Visualizing Content and
                Structure
• Visualizing large amounts of hierarchical
  information is a challenge
   – Users want both the content and the structure of
     hierarchical information
• Listings are good at showing the contents but not
  good at presenting the structure
   – Even with full path names listings do not help users in
     building a mental model of the structure
• Outlines show both structure and content
   – But require lot of display space
• Both listings and outlines require number of lines
  of display proportional to the number of nodes in
  the hierarchy
• Traditional tree drawings are good only for
  visualizing small trees
     Dept. of Computing Science, University of Aberdeen        5
                      Requirements
• Visualization scheme should utilise the 100%
  available display space
   – Traditional tree drawings utilise only 50% of available
     space
• Users should be able to control the properties of
  the visualization such as
   – Depth of the tree and
   – Content of the tree
• Visualization should be ‘readable’
   – Users should find it easy to scan the display
• Visualization scheme should extend gracefully to
  include additional properties of trees
   – As described later


     Dept. of Computing Science, University of Aberdeen        6
                               Treemaps
• Treemaps are novel displays of hierarchical
  information
   – Satisfy all the above requirements
   – Use 100% of the available display area
   – Algorithm for drawing treemaps is simple
   – No constraints on the maximum number of nodes in the
     tree
   – Variations of basic treemaps show trees with special
     properties (ordered trees etc)
• Historically treemaps were invented to display
  disk usage on a computer
   – Treemap layout displays all the files on the disk
     proportionate to their size (or any other property)
   – Users can interact with this layout (by dragging the
     mouse over a file) to obtain file details

       Dept. of Computing Science, University of Aberdeen   7
              Example: CS5561 folder
                    structure
                                                 CS5561




lectures                    practicals                          assessment              information




week1               week2                week3        week4        a1        internal                 external




           Dept. of Computing Science, University of Aberdeen                                                    8
          Nested Rectangles
CS5561




 lectures                   practicals                 assessment information

  Dept. of Computing Science, University of Aberdeen                        9
            Problems with nested
                 rectangles
• Not good for deep trees
   – Results into large degree of nesting of rectangles
• Adding labels not easy with long and lean
  rectangles
   – In the previous slide even at the third level it is hard to
     add text horizontally
• Leaner rectangles possible with increasing depth
  (or level)
• We want squares or near squares rather than
  rectangles
   – To reclaim space wasted in nesting offset
• Displaying large trees requires efficient use of
  available display area

     Dept. of Computing Science, University of Aberdeen            10
   Slice and Dice Algorithm
• Main idea is very simple
  – At each new level of the tree change the
    direction of partitioning of the rectangles
  – Hence the name slice and dice
• Imagine you start with a block of cheese
  – First slice it vertically
  – Then dice each piece from above horizontally



    Dept. of Computing Science, University of Aberdeen   11
             Example: Tree-map
     CS5561
                                 week1

                                                                internal
                                 week2                     a1
      lectures
                                 week3                          external

                                 week4


-Size of the display partition proportional to the size of the folder
-Other file attributes can be mapped to other attributes of the
partition such as color, texture etc
      Dept. of Computing Science, University of Aberdeen                   12
     Properties of Treemaps
• Aspect ratio
   – Max(width/height,height/width)
   – A square has an aspect ratio = 1
   – Slice-and-dice may produce rectangles with poor aspect
     ratio
• Readability
   – Ease of scanning the treemap for required information
   – e.g searching for a specific file
   – We stick to this informal definition
• Smoothness of change in the layout due to
  changes in the tree data
   – Files change on the disk all the time

     Dept. of Computing Science, University of Aberdeen       13
 Algorithms to Improve Aspect
       Ratio in Treemaps
• Several algorithms exist for improving
  aspect ratio
  – E.g. Map of the Market tool on
    SmartMoney.com uses clustered treemap
    method
  – Produces tree map with better aspect ratio
    (partitions closer to a square)
• But many of these algorithms produce
  treemaps with
  – Poor readability
  – Ordering information from the original data set
    not preserved

    Dept. of Computing Science, University of Aberdeen   14
           Ordered Treemaps
• Input tree contains ordered
  information
  – E.g. alphabetically ordered children
• Algorithms that maintain healthy
  aspect ratios and also preserve
  ordering information are available
  – You can look up the algorithms from the
    course information page
   Dept. of Computing Science, University of Aberdeen   15
           Quantum Treemaps
• The contents of a partition need not be
  always label strings
  – You could have images which need a minimum
    space to display
• Algorithms that ensure that the display
  space available in a partition is always a
  multiple of user specified quantum are
  available
  – You can look up the algorithms from the course
    information page

    Dept. of Computing Science, University of Aberdeen   16
                       Fisheye Views
• Address the problem of visualizing large
  amounts of any type of information (not
  necessarily tree information)
• Using zoom in/out is the common solution
  – Often available with geographic maps (e.g.
    Google Earth)
  – The zoom in operation offers a detailed local
    view (focus)
  – The zoom out operation offers a global view
    (context)
• Fisheye views offer an alternative way of
  displaying focus and context information
  – New Yorker’s view of the World’ drawing by
    Steinberg
  – http://en.wikipedia.org/wiki/Saul_Steinberg
    Dept. of Computing Science, University of Aberdeen   17
New Yorker’s
View of the World
-An example
Fisheye view




            Dept. of Computing Science, University of Aberdeen   18
        Natural Fisheye views
• Fish see details of the world directly above them
  but only a distorted view of the rest of the world
   – Due to light refraction
• Employees know local department heads but only
  the Vice Presidents of remote departments
• We all discriminate subfields of computing such as
  AI, DB and Networks but nor subfields of a
  remote discipline such as Psychology
• News papers carry several local news but only a
  few global news of great importance


     Dept. of Computing Science, University of Aberdeen   19
 Formal theory behind fisheye
            views
• At the heart of the fisheye views is the
  notion of degree of interest (DOI)
• DOI is composed of two parts
  – A priori importance (API)
  – Distance (D)
• DOIfisheye (x|.=y)=API(x)-D(x,y)
  –   X is the point for which DOI value is computed
  –   Y is the current point of focus
  –   DOI increases with API
  –   DOI decreases with D

      Dept. of Computing Science, University of Aberdeen   20
    Example 1: CS5561 folder
           structure
• Let us compute DOI for the CS5561 tree
  we have from the treemaps discussion
• Let the node a1 be the point of focus
• D(x,y) be the path length in the tree from
  x to y, dtree(x,y)
  – A very natural distance measure in trees
• API(x) be the path length between x and
  root of the tree, -dtree(x,root)
  – Negative sign shows that importance falls as
    you move away from the root

    Dept. of Computing Science, University of Aberdeen   21
              Example: CS5561 folder
                  structure (2)
                                                       CS5561
                                                        D=2
                                                       API=0
                                                       DOI=-2




lectures                       practicals                              assessment                      information
  D=3                            D=3                                      D=1                              D=3
API=-1                          API=-1                                   API=-1                          API=-1
DOI= - 4                       DOI= - 4                                 DOI= - 2                         DOI= - 4




 week1              week2                    week3          week4        a1                 internal                 external
  D=4                D=4                      D=4            D=4        D=0                   D=4                      D=4
API=-2              API=-2                  API=-2          API=-2     API=-2               API=-2                   API=-2
DOI= - 6           DOI = - 6                DOI= - 6       DOI = - 6   DOI = -2             DOI= -6                  DOI= - 6




           Dept. of Computing Science, University of Aberdeen                     Current                                       22
                                                                                   focus
        Example: CS5561 folder
            structure (3)
• There could be several ways of using DOI
  information to render fisheye views
   – DOI can be used for other purposes than just generating
     fisheye displays
   – Given some information, DOI helps to compute metrics to
     separate focus and context
   – In this sense fisheye views involve deeper significance
     than simply generating fisheye displays
• Let us use the size of the node in the display to
  indicate DOI
   – Use a threshold, k on DOI to select items for display


     Dept. of Computing Science, University of Aberdeen      23
     Example: CS5561 folder
         structure (4)

                                        CS5561



              lectures     practicals                         information

                                                 assessment


Threshold used is k=-4
All nodes with
                                                       a1
DOI>=k are shown
Size of the box is
proportional to DOI value
  Dept. of Computing Science, University of Aberdeen                        24
                             Summary
• Displaying large amounts of information on
  limited screen is a challenge
  – Hierarchical information can be displayed using
    treemaps
      • Slice-and-dice algorithm produces poor aspect ratios
      • Improving aspect ratio and retaining other properties
        such as readability, smoothness of updates, and
        ordering
  – Fisheye views can help to display any type of
    data
      • Present focus+context
      • Parts of the display is distorted


    Dept. of Computing Science, University of Aberdeen      25

				
DOCUMENT INFO
Categories:
Tags:
Stats:
views:7
posted:10/23/2011
language:English
pages:25