EECE 5642

Data Visualization

Table and Graph

Y. Raymond Fu
Professor
Electrical and Computer Engineering (ECE), COE
Khoury College of Computer Science (KCCS)
Northeastern University
 When to Use Tables vs. Graphs
• Use tables when
– The document will be used to look up individual values.
– The document will be used to compare individual values.
– Precise values are required.
– The quantitative information to be communicated involves more than
one unit of measure.
• Use graphs when
– The message is contained in the shape of the values.
– The document will be used to reveal relationships among multiple
values.

2 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Fundamental Variations of Tables
• Quantitative-to-categorical relationships
– Between one set of quantitative values and one set of categorical
subdivisions
– Between one set of quantitative values and the intersection of
multiple categories
– Between one set of quantitative values and the intersection of
hierarchical categories
• Quantitative-to-quantitative relationships
– Among one set of quantitative values associated with multiple
categorical subdivisions
– Among distinct sets of quantitative values associated with the same
categorical subdivision

3 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Quantitative-to-Categorical

4 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Quantitative-to-Quantitative

5 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Variations in Table Design
• Two fundamental types of structural variations
– Unidirectional—categorical items are laid out in one direction only (i.e.,
either across columns or down rows)
– Bidirectional—categorical items are laid out in both directions

6 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Summary

7 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Summary

8 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Table Evolution

9
 Table Evolution

10
 Table Evolution

11 Robust Spectral Ensemble Clustering, CIKM, 2016.

 Table Evolution

Adversarial Graph Auto-Encoders, 2018.

12
 Table Evolution

Adversarial Graph Embedding for Ensemble Clustering, IJCAI, 2019.

13
 Table Evolution

14 Knowing When to Look: Adaptive Attention via A Visual Sentinel for Image Captioning, CVPR, 2017
 Fundamental Variations of Graphs
• Graphical objects used to encode quantitative values
– Points
– Lines
– Bars
– Shapes with 2D area

15 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Points

Use both points and line for the same set

Quantitative data
of data is not redundant sometimes. The
line helps focus on the overall changes
and trend of the data.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

16 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Points
• Visibility problem

Stephen Few, Show Me the Numbers, Analytics Press, 2004

17 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Points
• Visibility problem

Stephen Few, Show Me the Numbers, Analytics Press, 2004

18 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Lines

Quantitative data Points for individual values and line for

overall trend of the data.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

19 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Lines
• Visibility problem

Stephen Few, Show Me the Numbers, Analytics Press, 2004

20 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Lines
• Use legend (additional text)
– Label
– Introduce
– Explain
– Reinforce
– Highlight
– Sequence
– Recommend
– Inquire

21 Stephen Few, Show Me the Numbers, Analytics Press, 2004

 Lines
• Visibility problem

Color and legend

Stephen Few, Show Me the Numbers, Analytics Press, 2004

22 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Visual Attributes for Categorical Data
• Position, Color, and Point Shape

Stephen Few, Show Me the Numbers, Analytics Press, 2004

23 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Visual Attributes for Categorical Data
• Fill Pattern and Line Style

Bad example!

Dizzying effect; moire vibration; junkchart

Distract your attention
Avoid using different line styles

Stephen Few, Show Me the Numbers, Analytics Press, 2004

24 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Seven types of relationships
– Nominal comparison
– Time series
– Ranking
– Part-to-whole
– Deviation
– Distribution
– Correlation

Stephen Few, Show Me the Numbers, Analytics Press, 2004

25 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Seven types of relationships
– Nominal comparison: Plus or minus; relative to; variance
– Time series: change; rise; increase; fluctuate; growl decline; trend
– Ranking: less than, equal to; smaller than
– Part-to-whole: percentage of total; share; accounts for X percent
– Deviation: difference; the degree to which sales of various products
differ from a specific one
– Distribution: frequency; range; normal distribution; how the data
spread over
– Correlation: relates to; increase/decrease with; varies with; caused by

Stephen Few, Show Me the Numbers, Analytics Press, 2004

26 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Nominal comparison
– No particular order
• Time series
– Quantitative data over time
• When to use time series?
– State the message verbally
• Change
• Rise
• Increase
• Fluctuate
• Grow
• Decline
• Decrease
• Trend

Stephen Few, Show Me the Numbers, Analytics Press, 2004

27 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Ranking
– Sorted by size
– Words and phrases suggest a
ranking relationship
• Larger than
• Smaller than
• Equal to
• Greater than
• Less than

• Part-to-whole
– Quantitative values
related to total value

Stephen Few, Show Me the Numbers, Analytics Press, 2004

28 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Deviation
– Show difference from a primary set of quantitative values
– Examples
• The degree to which actual worker productivity differs from target productivity
• The degree to which sales over time differ from sales at some specific time in the past
• The degree to which headcount for each month differed from headcount for the
previous month
• The degree to which sales of various products differ from sales of a particular product
– Use following units of measure
• Actual units (revenue dollars, number of employees, etc.)
• Ratios relative to the primary value
• Positive or negative ratios relative to the primary value
– Words and phrases express deviation relationship
• Plus or minus
• Variance
• Difference
• Relative to Stephen Few, Show Me the Numbers, Analytics Press, 2004
29 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Deviation
Show
difference to
a primary set
of values

Using percentages

Stephen Few, Show Me the Numbers, Analytics Press, 2004

30 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Deviation

Use zero as base! Bad example!

Stephen Few, Show Me the Numbers, Analytics Press, 2004

31 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Deviation combined with time-series

Stephen Few, Show Me the Numbers, Analytics Press, 2004

32 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Distributions

Histogram Frequency Distribution

(Frequency polygon)

Stephen Few, Show Me the Numbers, Analytics Press, 2004

33 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Multiple Distributions

Stephen Few, Show Me the Numbers, Analytics Press, 2004

34 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Correlations Scatter plot
– Two paired sets of
quantitative values vary in
relation to each other
– Words indicate a correlation
• Relates to
• Increases with
• Decreases with
• Changes with
• Varies with
• Caused by
• Affected by Employees’ heights and their salaries.
• Follows

Stephen Few, Show Me the Numbers, Analytics Press, 2004

35 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Correlations
– Scatter plot

Include a trend line to

highlight the correlation.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

36 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Relations in Graphs
• Correlations
– Correlation bar graph
– Paired bar graph

Stephen Few, Show Me the Numbers, Analytics Press, 2004

37 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Aspect Ratios
Same data; different aspect ratios

a = height / width

They all display

precisely the
same numbers,
but differ in
perceptual impact

Stephen Few, Show Me the Numbers, Analytics Press, 2004

38 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Line Scales

Highlight and emphasize!

Stephen Few, Show Me the Numbers, Analytics Press, 2004

39 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Bars

Quantitative data

Stephen Few, Show Me the Numbers, Analytics Press, 2004

40 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Bars
• Always begin the quantitative scale at zero

• Otherwise, will cause inaccurate comparisons.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

41 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 3D Bars
Avoid data objects with
3D depth: remember the
data-ink-ratio? The
addition of 3D to value-
encoding objects adds ink,
but no data.

But, The 3D depth added

to an entire graph
through the use of a third
axis is a different story.
This third axis may be
used for either a
categorical or a
quantitative scale.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

42 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Stacked Bars
Part-to-Whole graphs
relate parts of something
to the whole.

Sometimes, part-to-
whole relationships use
separate bars can work
better than stacked bars
to clearly display the
contribution of each part
to the whole.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

43 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Box & Whisker Plots

Example
Stephen Few, Show Me the Numbers, Analytics Press, 2004
44 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Multiple Box & Whisker Plots
• What are the insights
revealed in the display?
– On average, women are
paid less than men in all
salary grades.
– The disparity in salaries
between men and women
becomes increasingly
greater as one’s salary
increases.
– Salaries vary the most for
women in the higher
salary grades.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

45 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Pie Charts

Again, part-to-
whole relationships
use separate bars
can work better
than pie charts to
clearly display the
contribution of
each part to the
whole.

Stephen Few— avoid pie charts. It communicates

poorly. Our visual perception is not designed to
accurately assign quantitative values to areas. We
have an even harder time when the third
dimension of depth is added.

Stephen Few, Show Me the Numbers, Analytics Press, 2004

46 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Bubble Graphs

Create a Bubble Chart with Multiple Series of Data

https://www.youtube.com/watch?v=bUv_pEO9ZFc
Gapminder Foundation, http://www.gapminder.org/
47 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Basic Data Analysis and Interpretation

https://www.youtube.com/watch?v=dmw4xnRTnU4

48
 Exercise
Which visual encoding objects display each relationship best?

Stephen Few, Show Me the Numbers, Analytics Press, 2004

49 Courtesy of Prof. Hanspeter Pfister, Harvard University.
 Answer

50
Stephen Few, Show Me the Numbers, Analytics Press, 2004
 Proposal Presentation
• 6~8 minutes for each person/team
• Online students could send me your slides by adding notes for each slide
• Required Content
– Project title
– Team members’ names, affiliations and emails (one or two additional members)
– The project option you choose (demo or tool)
– Motivations of the project
– Real‐world applications
– Data source and background
– Tools and programming languages used in the project
– Contributions of the work (the work by the authors)
– Novelty of the work (optional)
– Visualization techniques (need to present details)
– Division of work for each team member
– Challenges and solutions
– Timeline and schedule

51
 Plot Evolution

52
 Plot Evolution

53
 Plot Evolution

54
 Some Good Examples

55 YellowFin and the Art of Momentum Tuning, SysM

 Some Good Examples

56 YellowFin and the Art of Momentum Tuning, SysM

 Some Good Examples

Metric Learning on Healthcare Data with Incomplete Modalities , IJCAI’19

57
 Some Good Examples

https://arxiv.org/abs/1904.08587
58