CS 4460 Exam 1

What is visualization?

The use of computer-supported, interactive visual representation of data to amplify cognition.

Often thought of as a process of making a graphic or an image, but is really in support of cognitive processes

Purpose of visualization is insight, not pictures

Two main purposes/uses of InfoVis

1) Analysis: understand, solve problems, answer questions, performing tasks

2) Presentation: use vis to communicate ideas, influence, explain, persuade

When to use InfoVis

More useful in exploratory data analysis

Let yourself discover, but also explain and answer specific questions ("...highest value when it forces us to see what we were not expecting)

When not to use InfoVis

When you know exactly what questions to ask --> other techniques that give explicit answers may be better (statistics, data mining, ML, etc.)

Analysic tasks for InfoVis

Search (ok) - finding a specific piece of information

Browse (better) - look over or inspect something in a more casual manner, seek interesting information

Analysis (best) - comparison/difference, outliers, extremes, patterns

Key benefits of visualization

Facilitating awareness and understanding

Helping to raise new questions and supply answers

Generating insights

Telling a story faithfully and making a point

Visual analytics and its main components

Combines analytical reasoning and interactive visualizations to help effectively understand large and complex data sets

Main components are:

1) interactive visualization

2) Analytical reasoning

3) Computational analysis

Nominal data type

Categorical

e.g. gender

Ordinal data type

Ordered set, obeys < relation

e.g. freshman, sophomore, junior, senior

Ordinal = ordered

Quantitative data type

Can do math, equal intervals

e.g. distance, weight

Visual mark effective

Nominal ones will have:

- spatial region

- color hue (different colors as in like red, blue, green, etc)

- motion

- shape

Why? Cause they're categories. There's not really a sense in having markers that are scale-like. So like one of those lines with the circle you pull up and down. Or different size squares/dots. You have totally different colors to differentiate bars on a bar graph for instance.

Ordinal & Quantitative:

- Using scale things

- depth (3D position)

- area (2D size)

- tilt of line/point

- curvature

- volume

Cause these are all about order or different scales

Data marks

Visual primitives in 2d/3d space

Graphical properties of data marks encode data values

(Size, shape, color (HSV), orientation, texture, border, thickness)

Other names: visual encodings, graphical encodings, visual glyphs

What are information visualizations made up of?

Data marks placed inside of visual metaphors and mapped to data values

What are data items, data attributes, and data marks?

Data items: an entity in the data set (aka: data cases, rows)

Data attributes: a value measured for all data cases (aka: data features, dimensions, variables). Columns

Data marks: visual encodings, visual marks, graphical encodings

(colors, length of bars, etc.)

Example:

Data items: different customers

Data attributes: different things the customers do

Data marks: graphical properties (color, thickness, orientation, etc.)

Univariate data

One variable

Data items shown on one dimension and the value (only attribute) on the other

Eg: line chart, bar graph

Bivariate data

Two variables

Scatterplot is common, we want to see relationship between the 2 variables

Trivariate data

Three variables

3d scatterplot is possible (but not great)

Alternative is to still use 2D, but have visual property represent third variable (ex. size)

Alternative is to represent each variable in its own explicit way

Hypervariate data

Greater than three variables

Often the more interesting (but challenging) ones

Could use additional data mark properties to encode additional data variables

Few's Selection and Design Process

1) Determine message and identify data

2) Table, graph, or both needed to communicate message

3) Determine best means to encode values

4) Where to display each variable

5) Determine best design (legend, range, scale, grid, text etc)

6) Which particular data be featured and how

Basically you need to understand what's going on, decided what type of chart/graph/table thingie to use, decide how to show the values, decide where to display variable (something that can be measured), determine the InDesign or Illustrator prettiness, what feature to be special

Parallel-coordinates plot

Plot it point by point per line (D1 x V1 - V5, D2 x V1 - V5 etc)

Disadvantages of parallel coordinate graphs

Lots of clutter (especially if there's too many different lines)

Can only see trends between two adjacent point (like per column)

It's hard to trace one data item (to fix this, you can add interactivity by rearranging them)

Star plots

Space out the n variables at equal angles around a circle

Each spoke (point of the star) is a variable value

Think of a sports manga and the stats for the characters/players

Perceptual processing

Seeks to better understand visual perception and visual information processing

This is how we see, therefore the "inputs" to the brain for visualization

Understand how we see, basically

Colin Ware's Model for Perceptual Processing

Stage 1: early, parallel detection of color, texture, shape, spatial attribute.

Automatically, rapid, aka (pre-attentive processing)

Stage 2: serial processing of object identification (using memory) and spatial layout action. Sequential, goal-directed, slow serial processing

So in other words, we detect color and forms first. Then we actually realize that what it is.

Preattentive processing

The mental shortcut that your mind can/and does use

Done in parallel by low-level vision system

It's fast and focused attention is unnecessary

That's why it's called "preattentive". Before attentive.

Works well for:

- Target detection

- Boundary detection

- Counting

- Spotting different shapes

- Spotting different hues

Basically doing basic/implicit (?) tasks

Not good for:

- Conjunction of features (space and hue) causes a problem. So if you have shape and shade of color at the same time, comprehending that with this is difficult.

- A potential danger would be calling attention to something that isn't very important (using too much of bright red)

Gestalt principles

1) Proximity: we group together proximate objects

2) Similarity: we group similar objects together

3) Common fate: we group together objects seen to be moving in the same direction (having a common fate)

4) Symmetry: elements that are symmetrical to each other, they tend to be perceived as a unified group

5) Closure: we try to see collections of objects as creating a larger, more complete object (the Pacmans creating a triangle)

6) Continuity: we separate (parse) overlapping objects to give them a smooth interpretation (fill in the blanks). Eye moves from one object to another. A flow.

7) Connectness: connected visual elements are seen as a whole

8) Figure/ground: think of optical illusions (face/vase image)

Importance of gestalt principles Info Vis

Using visual structure helps reinforce the underlying logical structure of data

Parallel sets

- Continuous axes in parallel coordinates fare replaced with boxes in parallel sets

- Parallel sets is a visualization method with the parallel coordinates layout but uses frequency-based representation

Parallel sets also have interaction --> users can create new classifications

This is the graph on Titanic

Star plots vs star coordinates plots

Star plots --> a data case is a shape

- star coordinates

- a data case is a point. Rather than represent point as polyline, just accumulate values along a vector parallel to particular axis

- data cases with similar values will lead to clusters of points

**In a star plot, an entire shape is a data item, but in a star coordinates plot, a data item is a point

Star plot: sports stat graph

Star coordinate plot: the stupid quiz we did in class where we did a bunch of points on a five line star thingie

Dust and magnet visualization

Iron bits (data cases) are drawn towards magnets (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

e.g. the cereal dust and magnet example. Magnets are sugar and sodium and the dust are cereal types

Graphical excellence

Design aesthetics: consists of complex ideas communicated effectively with clarity, precision and efficiency

Graphical integrity: telling truth about the data

Chart integrity principles

Baseline: where is 0 on the chart if it's even there

Scale: are the data values visually accurate amongst all data cases?

Context: are the visuals necessary to the understanding of the graph

Measuring misrepresentation

Visual attribute value should be directly proportional to data attribute value

Lie factor = size of effect shown in graphic / size of effect in data

Design aesthetics

- Definition: set of principles to help guide designers

- Show the data!!!!

- Maximise data-ink ratio:

1) Erase non-data-ink

2) Erase redundant data-ink

3) Revise and edit

Basically don't have things that aren't data and make sure there's no redundancy

Data ink ratio = data ink / total ink used in graphic

= proportion of graphic's ink devoted to the non-redundant display of data-information

Avoid chartjunk

Chartjunk are extraneous visual elements that distract from message.

It's all about PURPOSE. Some embellished charts are just as good on interpretation accuracy and even recalled better weeks later.

Utilize multi-functioning graphical elements (macro/micro readings)

- Graphical elements that convey data information and a design function

- Stem and leaf displays

Use small multiples

Repeat visually similar graphical elements nearby rather than spreading them apart

Like if you use a bunch of different shapes or colors it'll be really confusing

Utilize narratives of space and time

Tell a story of position and chronology through visual elements

Ex: the life of a beetle visual (so the beetle growing up)

Content is KING

Quality relevance, and integrity of content is fundamental

Make visual design reflect the analysis task

Dos and Don't from Google

- Bar graphs for quantity and change over time and pie charts for proportion

- Bar charts with subtle rounded corners (we want to make sure the top of bar measures precisely the bar's length)

- Use a combination of color highlights and neutral colors to provide contrast and emphasis

Webstack intro

HTML: Provides basic sturcutre of web page, enhanced and modified by CSS, JavaScript

CSS: for styling, controls presentation, formatting, and layout

JavaScript: controls the behavior of different elements on a web page

HTML, CSS, SVG

- Has coordinate system (x and y)

- Think of HW2 where we made the bar, graph, chart etc

- scale(x,y)

- Translate(x,y)

- rotate(45)

- .

Task taxonomies

Structural decompositions of concepts to help us understand categories and their relationships

So you break down concepts to understand categories and their relation to each other

Sheniderman's mantra

Task x data type taxonomy --> what people do with visualization

- Overview: helps present overall patterns, assists users with navigation and search, orients activities, visualizations should start with an overview

- Zoom and filter: focus on area of interest using interactive features

- Details on demand: progress deeper to find and focus on details of interest

**Overview and details can be combined through interaction or by using different portions of the screen

10 low-level tasks in the primary tax taxonomy (Amaer, Eagan, & Stasko)

1) Retrieve value: given a set of specific cases, find attributes of those cases

e.g. How long is the movie Gone with the Wind?

2) Filter: given some concrete conditions on attribute values, find data cases satisfying those conditions (and hide others)

e.g. What comedies have won awards?

3) Compute derived value: given a set of data cases, compute an aggregate numeric representation of those data cases

a. Aggregate function: a function that creates a numeric for a set of representation for a set of data cases (e.g. average, sum)

b. e.g. What is the average calorie content of Post cereals?

Find extremum

Find data cases possessing an extreme value of an attribute over its range within the data set

e.g. what is the car with the highest MPG?

Sort

Given a set of data cases, rank them according to some ordinal metric

a. e.g. order cars by weight

Determine range

Given a set of data cases and an attribute of interest, find the span of values within the set

e.g. what is the range of film lengths (i.e. 1 - 3 hr)

Characterize distribution

Given a set of data cases and a quantitative attribute of interest, characterize the distribution of that attribute's value over the set

e.g. what is the age distribution of shoppers

Find anomalies

Identify any anomalies within a given set of data cases with respect to a given relationship or expecation

e.g. statistical outliers

e.g. are there any outliers in protein?

Cluster

Given a set of data cases, find clusters of data items with similar attribute values

e.g. are there groups of cereals w/ similar fat/calories/sugar

Correlate

Given a set of data cases and two attributes, determine useful relationships between the values of those attributes

e.g. is there a correlation between carbs and fats?

Brushing and linking

- Popular mechanism to coordinate two visualizations

- Selection in 1 view highlights corresponding data items in second view

So you select your brush in one area and it hightlight same points in other views

Interactive Dynamics (part interaction, part task)

Data and view specification

View manipulation

Process and Provenance

7 techniques for showing overview + detail

1) detail-only: single window with horizontal and vertical panning

2) single window with zoom and replace: global view with selectable zoom area which then becomes entire view

3) single coordinate pair: combined display of the overview and local magnified view (separate views)

4) Tiled multilevel broswer: combined global, intermediate, and detail views that do not overal

5) Free zoom and multiple overlap: overview presented first; user selects area to zoom and area in which to create detailed view

6) Bifocal magnified: magnified glass zoomed image floats over overview image

7) fish-eye view: magnified image is distroted so that focus is at high magnification, periphery at low

Focus + context views

Key difference to overview and detail: overview and detail are combined into a single display

Fisheye terminology

Most well-known focus + context view

- Focal point: some item, coordinate, or position users focus on

- Level of detail: some intrinsic value or quantity on each data element

- Distance from focus: calculation of how far each data item is from the focal point

- Degree of interest function: function that determines how items in display are rendered

- Degree of interest = level of detail - distance from focus

Degree of interest function types

- Continuous: smooth interpolation away from focus

- Filtering: past a certain point, objects disappear

- Step: level or regions dictating rendering

- Semantic changes: objects change rendering at different levels

Interaction

- User goals and tasks carried out through interaction with visualization

- Interaction transforms viewers into active participants in the analysis process

- Interaction moves visualization from one state to another

Mental models

We construct mental models based on our experiences during out life. It's constantly created, adjusted

Sensemaking loop

Overview: provides empirical (observed) descriptions of the cognitive processes involved in intelligence analysis

Intelligence analysis is intended to gain evidence and support from schemas/spatial structures (ordered lists, hierarchical organizational charts, etc) that hold tacit information

Bottom-up approach:

- Search and filter

- Read and extract

- Schematrize

- Build case

- Tell story

Top-down approach:

- Reevaluate

- Search for support

- Search for evidence

- Search for relations

-Search for information

Information foraging tasks

1) Exploring: recall, increasing span of new info

2) Enriching: narrowing to smaller set of items

3) Exploiting: generate inferences, notice pattern

Data frame model

- Process of building a mental model when performing analysis on a visualization

- Seek a frame --> elaborate the frame --> question the frame --> preserve the frame --> compare frames

1) Seek a frame: establishing new relationships that allow you to explain a collection of data

2) Elaborate the frame: seek data to fill our and confirm existing frame and extending the frame to "fill slots" with data that you expected, then incorporate back into collection of frames

3) Questioning the frame: found data that violates an expectation that you have about frame

4) preserving frame: explain the anomaly in data the forced you to question and understand the error in data (frame holds when understood).

5) comparing frame: act of sharpening and refinding the distinction between frames to clarify the boundaries/definitions of both

While there are loops, many of these tasks happen quickly and in parallel

process can happen even without any interaction with a system

Defining insight

1. A product or result:

a. complex

b. deep

c. qualitative

d. unexpected

e. relevant: dependent on the domain

2. A process

a. considering prior knowledge

Spacial organization of information is a good method to use visualization in direct support of sensemaking