The Table Lens: Merging Graphical and Symbolic Representations in an Interactive Focus Context Visualization for Tabular Information Ramana Rao and Stuart K. Card (pgs. 343-349)

1
The Table Lens Merging Graphical and Symbolic
Representations in an Interactive FocusContext
Visualization for Tabular InformationRamana Rao
and Stuart K. Card(pgs. 343-349)
2
Introduction

• Uses a focuscontext (fisheye) technique.
• Allows display of crucial label info. and
  multiple distal focal areas.
• Created because of limitations in traditional
  spreadsheet applications.

3
Comparison to Traditional Spreadsheet Application

• The size of the information set which users can
  display limits the ability of users to address
  complex problems.
• Excel can only display a max of 660 cells on a
  19 display.
• Table Lens can display 30-100 times as many
  cells, depending on task.

4
Other Advantages of Table Lens

• Merges graphical representations directly into
  the tables (automatic plots integrated into
  cells).
• Allows humans to spot patterns and features much
  easier.
• Makes exploration of data much more interactive
  and natural.

5
FocusContext Technique

• Supports visualizing an entire information
  structure at once or zooming in on specific
  items.
• Evolved from other similar techniques
• Bifocal Display
• Furnass Fisheye
• Perspective Wall
• Document Lens

6
FocusContext Technique

• Table Lens mutates the layout of a table without
  bending any rows or columns.
• Distorts based on cell boundaries.
• Distortion in each of the two dimensions is
  independent from the other.
• Allows for easy horizontal and vertical eye
  scanning.
• Enables label display, multiple focal areas, and
  multiple focal levels.

7
FocusContext Technique

• 4 types of cell regions are created by the
  distortions on the two axis
• Focal
• Row focal
• Column focal
• Nonfocal
• Focal data is textual, while nonfocal data is
  graphical.

8
FocusContext Technique
9
The Distortion Function

• Core based on Degree of Interest (DOI) function.
• Maps from an item to a value that indicates the
  level of interest in the item.
• Used to control how available space is allocated.
• Transfer function
• Maps uniformly distributed cell addresses to
  interest-warped physical locations.

10
The Distortion Function

• Similar to Bifocal Display, except there are 2
  independent distortions on vertical and
  horizontal.
• Contrasts with Perspective Wall and Document Lens
  which map z-surfaces over a flat plane.

11
Interactive Manipulation of the Focus

• 3 main types
• Zoom
• Changes amount of space without changing number
  of cells.
• Adjust
• Changes amount of contents without changing the
  amount of space (size of focal area).
• Slide
• Changes location of entire focus area.

12
Interactive Manipulation of the Focus

• Visualizing the DOI function as the three
  manipulations are performed

13
Interactive Manipulation of the Focus

• A 4th type of manipulation is a coordinated
  adjust and zoom (adjust-zoom).
• Used to increase or decrease the number of focal
  cells w/o affecting their size.
• Multiple focal levels creates a complex design
  space, when individual focal areas are formatted
  differently from each other.

14
Graphical Mapping Scheme

• Designed for most common type of table
• Cases-by-variable array (Relational database)
• Cases are the rows and values of the various
  variables (across cases) are in the columns.
• Number of different types of graphical
  representations (presentation types) are used.
• e.g. Text, color, shading, length, and position

15
Graphical Mapping Scheme

• Presentation type determined by 6 factors
• Value
• Value Type
• Region Type
• Cell Size
• User Choices
• Spotlighting

16
User Interface

• Small number of key commands, but most of the
  time mouse gestures can be used exclusively.
• Left mouse button click is for touching.
• Left mouse button held down is for grasping.
• Right mouse button brings up menu for selecting a
  focal area and spotlighting it, etc.

17
User Interface

• Grasping control points on a focus allows
  adjust-focus manipulation.
• Touching a context region will slide the current
  focus there.
• You can also grasp and drag a focus to a
  different region.

18
User Interface

• Columns can be rearranged by grasping the column
  label.
• Columns can be sorted by clicking on the column
  label. Click a second time on the label to
  toggle between ascending and descending orders.

19
Critique

• Strengths
• Supports effective interaction with very large
  tables
• Merges graphical representations directly into
  the process of table visualization and
  manipulation
• Efficient display of cell values
• Easily detect patterns and features, and find
  relationships between variables.
• Good for tabular and proportional data.

20
Critique

• Weaknesses
• Not useful for non-tabular data.
• Large number of attributes (many columns) may
  make scanning variables difficult.

21
Table Lens as a Tool for Making Sense of
DataPeter Pirolli and Ramana Rao(pgs. 597-615)
22
Introduction

• Compares performance of the Table Lens and Splus
  (a more traditional, command-based graphical
  tool) in an Exploratory Data Analysis (EDA) task.

23
Introduction

• Focus on 2 typical EDA (Sensemaking) tasks
  involving multivariate datasets
• Assessing a batch of data and finding the
  features of each variable.
• Finding lawful relations among a set of observed
  variables.

24
Sensemaking

• Refers to activities in which external
  representations such as texts, tables, or figures
  are interpreted into some meaningful manner.
• The data is basically summarized and abstracted
  differently.

25
Sensemaking

• Learning Loop Complex
• Search for representations to capture important
  regularities.
• Information is encoded into the representation.
• Ill-fitting residue information leads search for
  more accurate, informed representation.
• Representation drives search for information.

26
Sensemaking

• Human problem solver is viewed as an
  information-processing system with a problem.
• Data Fit Residual
• Example Developing an equation that predicts a
  dependent variable value based on an independent
  variable value.

27
Representative EDA Tasks

• Involves uncovering (like a detective)
  regularities, irregularities, and relationships
  between variables.
• This paper focuses on the two tasks mentioned in
  the introduction.

28
Representative EDA Tasks

• First task involves browsing the values of each
  variable (i.e. scanning columns in Table Lens).
• Assess the following
• Batch symmetry
• Spread
• Outliers
• Clusters
• Multiple nodes

29
Representative EDA Tasks

• Second task is an iterative process
• First step is to find a candidate variable that
  highly correlates with another variable of
  interest.
• Find possible effects of additional independent
  variables to further explain the residual.

30
EDA Tools Comparison

• Splus and Excel are more richly featured than
  Table Lens.

31
EDA Tools Comparison

• Table Lens
• Classifying the shape and skew is a skill that is
  somewhat different with a batch of values than it
  is for a histogram.
• Sorting is first step in correlation search.
• Scan across columns to identify other columns
  which show similar trend to sorted column.
• Column can be formatted to focus on 5 value
  summary.

32
EDA Tools Comparison

• Splus
• User invokes a brush tool which is a matrix of
  scatterplots.
• Histogram can also be displayed for each
  variable.
• Batch assessment can be done by looking at
  histograms.
• Can also invoke a stem-and-leaf plot or box plot.
• Cross variable correlation done by scanning row
  for scatterplots with strong trends (lines).

33
GOMS analysis

• Explore space of content and possible courses of
  actions available before exploiting them.
• Breakdown of actions required for each step of
  task.

34
Results of GOMS analysis

• It may be more efficient to iterate through a
  batch of variables in Table Lens.
• Table Lens achieves comparable performance with
  Splus, but it is also a much simpler interface
  that is easier to learn.

35
Design Refinement

• Boxplots
• Horizontal extent of a Table Lens column
  represents a coordinate system.
• Boxplot could be superimposed
• Rows representing outliers can have
  different-colored bars.
• Direct Manipulation
• ladder of powers
• Tames non-linearities
• Create manipulation that allows transforming the
  column interactively.
• Slider or control points on curve.

36
Design Refinement

• Variable permutations
• More correlated variables (columns) on each side
  of the sorted variables are automatically brought
  close together.
• Fit marks and residual curves
• Allows user to see how closely data in one
  variable really correlates with data in another
  variable.
• Allows user to see shape of residual data and
  then separate this data into a separate column.