Main issues in InfoVis

InfoVis, Universidade de Aveiro Beatriz Sousa Santos, 2017/18

https://www.youtube.com/watch?v=Soq428T8mDY

Why have a human in the decision-making loop?

Why have a computer in the loop?

Why use an external representation?

Why depend on vision?

Why show the data in detail?

Why use interactivity?

What is the design space of visualization idioms?

Why focus on tasks?

Why are most designs ineffective?

What resource limitations matter?

How can better be measured?

2

(Munzner, 2014)

Interviews with Netflix Data Scientists

:

https://classroom.udacity.com/courses/ud404/lessons/9239573934/concepts/

91687840320923

3

4

https://classroom.udacity.com/courses/ud404/lessons/9259930027/concepts/

91687840330923

Interviews with Netflix Data Scientists

How can we produce a Visualization?

5

• There are no “recipes” to chose adequate Visualization techniques

• There are principles (derived form human perception and cognition)

paradigms (examples resulting form past experience)

and many methods

• To obtain efficacy it is fundamental:

– a correct definition of goal and user tasks

– apply adequate methods and evaluate

in several iterations until the goals are satisfied …

A search space metaphor for Visualization design

• Only a very small number of possibilities are reasonable …

most are ineffective

Consider multiple alternatives and then choose the best!

6

(Munzner, 2014)

Framework for analysing Visualization use

What ?

Why?

How?

7

What data the user sees

Why the user intends to use a Vis tool (task)

How the visual encoding and interaction idioms are constructed

Simple Vis tools may be analysed as an instance;

Complex tools may require analysis in terms of a sequence of instances

(Munzner, 2014)

Data Characteristics

Beatriz Sousa Santos, 2017/18

Universidade de Aveiro

Departamento de Electrónica,

Telecomunicações e Informática

InfoVis, Universidade de Aveiro

10

• Data may have a lot of different forms and there are many techniques and systems to visualize them

• A data classification is important to:

- predict what visualization techniques are adequate

- make easier the communication about the data

- allow a more systematic approach to Visualization

….

What: Data Abstraction

• Four basic dataset types:

– Tables

– Networks

– Fields

– Geometry

• Five basic datatypes

– Items

– Attributes

– Links

– Positions

– Grids

11

Categorical

Ordered Ordinal

Quantitative

12

• Data representation level:

- Qualitative (or categorical)

- Quantitative (or numeric)

• Data nature:

- Continuous

- Discrete

• Measuring scale:

- Nominal

- Ordinal

- Interval

- Ratio

(Spence, 2007)

13

• Examples of measuring scales and types of data:

– nominal --> car brands, gender, animal species…

– ordinal --> week days, preferences, levels measured in a Likert-type scale

– Interval --> date, IQ, temperatures in ºC

– Ratio --> temperatures in ºK, weight, height

• The ratio scale represents the highest level of representation, has a non-arbitrary zero (unlike the interval scale)

• This is a general classification and might be used to select the statistical methods to use with the data

14

• Consider a data set with three columns:

latitude longitude d

• Which is the most adequate way to visualize these data?

• If d is depth, probably the selected visualization technique

will involve interpolation (ex: contours)

• If data represent location and the number victims of traffic accidents, interpolation and contours do not make sense

Know the data structure is not enough

It is necessary to know the phenomenon behind the data!

Model, structure and format of the data

to Visualize

• To understand the data it is necessary to know:

– Semantics – real world meaning

– Type – structural or mathematical interpretation

• Sometimes types and semantics may be inferred by observing the data

• Often they must be provided as metadata

16

Basil 7 S Orange

What is the meaning of this?

17

• What are the main issues in a InfoVis course?

• We present a typical example:

selecting a car to buy

• An instance of the general task of selecting one object from among many on the basis of its attributes

• An associated and crucial subtask is gaining insight into a collection of data

• An essential component of data mining and decision support

18

Example: selecting a car to buy

This type of problem is frequently characterized by lack of precision, involving:

- requirements difficult to specify as “nice looking”, “sporty” and “affordable”

- criteria of which the buyer is unaware, but influencing choice

The problem is often formulated as it is being solved!

19

Example: selecting a car to buy

The Data can be presented in tabular form

Rows - objects

Columns - attributes

However, tables can be of limited help, specially if there are

- a lot of rows

- 10 or more columns

An interactive rearrangement according to some criterion can be very helpful

(Spence, 2007)

20

10 - 12 12 - 14 16 - 18£kPrice

• A bargram of the price ranges associated the collection of cars is a semi-

quantitative representation useful when the price range is sufficiently informative

• The width of each bargram is proportional to the number of the objects (cars)

within each range

• A quick glance at the bargram helps the user in forming a useful mental model

• The bargram provides an overview of an attribute (price) and population

(Spence, 2007)

Example: selecting a car to buy using visualization

21

One unquantifiable attribute of a car

(appearance) may attract the buyer’s attention

The price of a car selected via its image is indicated by color coding both on

the image and on the corresponding icon (Spence, 2007)

(Spence, 2007)

Then, it would be natural to seek detail

The position of the car on the price bargram

can be indicated by an icon

The relation between a car’s image and the corresponding price range may be

shown by color coding

22

The position of a selected car is indicated on all attribute bargrams by color (Spence, 2007)

10 - 12 12 - 14 16 - 18£kPrice

MPG 30 35 40

• Price is not usually the sole criterion for choosing a car

• Multiple attributes of interest can be shown as bargrams with associated icons

• Again color can be used to relate the car image to the attribute ranges

24

The interactive selection of a bargram range (£12-14k) identifies four cars whose

price falls within that range (Spence, 2007)

10 - 12 12 - 14 16 - 18£kPrice

MPG 30 35 40

12 - 14

• The buyer may prefer to indicate an affordable price range interactively

• This corresponds to a focusing operation

25

• Subsequent interactive selection of a MPG range identifies only cars

which satisfy both requirements

10 - 12 12 - 14 16 - 18£kPrice

MPG 30 35 40

12 - 14

(Spence, 2007)

26

• A very powerful technique, brushing, allows to know which purple icon above

a bargram corresponds to a given framed car

• “The idea of linking and brushing is to combine different visualization

methods to overcome the shortcomings of single techniques. Interactive

changes made in one visualization are automatically reflected in the other

visualizations. Note that connecting multiple visualizations through

interactive linking and brushing provides more information than considering

the component visualizations independently”

D. A. Keim, Information Visualization and Visual Data Mining, IEEE

Transactions on Visualization and computer graphics, 2002

It allows users to better understand elements or variables across multiple views

27

http://www.infovis-wiki.net/index.php?title=Linking_and_Brushing

28

• There may be insufficient room to display all relevant attribute bargrams

• One solution is to scroll bargrams through a window

10 - 12 12 - 14 16 - 18£kPrice

MPG 30 35 40

12 - 14

(Spence, 2007)

Example: brushing

29

https://spotfire.tibco.com/resources/product

-demonstration-interactive/performance-

attribution-analytics

30 https://spotfire.tibco.com/resources/product-demonstration-

interactive/expense-analytics

Example: brushing

Example: Scrolling

31

32

A scrolling action determines which bargrams are fully displayed (Spence, 2007)

30MPG 35 40

10 - 12 12 - 14 16 - 18£kPrice 12 - 14

30

age

rec’n

HP

make

color

taxed

cond’n

• An alternative presentation approach is to reduce the vertical size

of many bargrams by suppressing range values while keeping the icons

33

If a selected Price range is the only one of interest,

other irrelevant detail can be suppressed (Spence, 2007)

Price £k 16 - 18

30 35 40MPG

• A filtering option could ensure the presentation of only data associated with cars

within an interesting attribute range

• The suppression of data might reduce the cognitive effort required from the user

• A disadvantage would be the loss

of valuable context information

37

Interaction with data governed by high-order cognitive processes:

- Representation

- Presentation

- Interaction

DATA

PERCEPTION

INTERPRETATION

REPRESENTATION of data

PRESENTATION of the

represented data

INTERACTION to select

the required v iew of data

The scope of this book

HIGHER-ORDER

COGNITIVE

PROCESSES

Internal modelling

Strategy formulation

Problem (re)formulation

Evaluation of options

Decision making

etc.

(Spence, 2007)

40

Main bibliography

• Spence, R., Information Visualization, Design for Interaction, 2nd ed.,

Prentice Hall, 2007

• Munzner, T., Visualization Analysis and Design, A K Peters, 2014

• Mazza, R., Introduction to Information Visualization, Springer, 2009

Acknowledgement

The author of these slides is very grateful to Professor Robert Spence as he

provided the electronic version of his book figures