The Table Lens Merging Graphical and Symbolic Representations

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)

Introduction l Uses a focus+context (fisheye) technique. l Allows display of crucial label info. and multiple distal focal areas. l Created because of limitations in traditional spreadsheet applications.

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

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

Focus+Context Technique l Supports visualizing an entire information structure at once or zooming in on specific items. l Evolved from other similar techniques: l Bifocal Display l Furnas’s Fisheye l Perspective Wall l Document Lens

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

Focus+Context Technique l 4 types of cell regions are created by the distortions on the two axis: l Focal l Row focal l Column focal l Nonfocal l Focal data is textual, while nonfocal data is graphical.

Focus+Context Technique

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

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

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

Interactive Manipulation of the Focus l Visualizing the DOI function as the three manipulations are performed: DOI DOI Zoom Adjust Slide

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

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

Graphical Mapping Scheme l Presentation type determined by 6 factors: l Value Type l Region Type l Cell Size l User Choices l Spotlighting

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

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

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

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

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

Table Lens as a Tool for Making Sense of Data Peter Pirolli and Ramana Rao (pgs. 597 -615)

Introduction l Compares performance of the Table Lens and Splus (a more traditional, commandbased graphical tool) in an Exploratory Data Analysis (EDA) task.

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

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

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

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

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

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

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

EDA Tools Comparison l Splus and Excel are more richly featured than Table Lens.

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

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

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

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

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

Design Refinement l Variable permutations l l 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. l Allows user to see shape of residual data and then separate this data into a separate column. l