cis 467/602-01: data visualizationdkoop/cis467/lectures/lecture11.pdf · simpler themeriver layout...

CIS 467, Spring 2015

CIS 467/602-01: Data Visualization

Tables

Dr. David Koop

Assignment 2• http://www.cis.umassd.edu/

~dkoop/cis467/assignment2.html • Plagiarism on Assignment 1 • Any questions?

2CIS 467, Spring 2015

http://www.cis.umassd.edu/~dkoop/cis467/assignment2.html

Recap (Interaction)• Important in visualization; it enables many tasks • Types of interactions:

- Feedback/Animation - Storytelling - Selection - Details/Highlighting - Navigation

• Concerns: - Change Blindness - Latency - Human Effort


Arrange Tables

Express Values

Separate, Order, Align Regions

Axis Orientation

Layout Density

Dense Space-Filling

Separate Order Align

1 Key 2 Keys 3 Keys Many KeysList Recursive SubdivisionVolumeMatrix

Rectilinear Parallel Radial

Recap (Tables)


[Munzner (ill. Maguire), 2014]

Recap (Scatterplots)• Data: two quantitative values • Task: find trends, clusters, outliers • How: marks at spatial position in

horizontal and vertical directions • Scalability: hundreds of points • Correlation: dependence between

two attributes - Positive and negative correlation - Indicated by lines

• Coordinate system (axes) and labels are important!


0 50 100 150 200 250 300 350 400 4500

50

100

150

200

250

300

350

400

450

Prices in 1970

Pric

es in

198

0

Fish Prices over the Years

Separate, Order, and Align: Categorical Regions• Categorical: =, != • Spatial position can be used for categorical attributes • Use regions, distinct contiguous bounded areas, to encode

categorical attributes • Three operations on the regions:

- Separate (use categorical attribute) - Align - Order

• Alignment and order can use same or different attribute


(use some other ordered attribute)

List Alignment: Bar Charts• Data: one quantitative attribute, one

categorical attribute • Task: lookup & compare values • How: line marks, vertical position

(quantitative), horizontal position (categorical)

• What about length? • Ordering criteria: alphabetical or

using quantitative attribute • Scalability: distinguishability

- bars at least one pixel wide - hundreds


100

75

50

25

0

Animal Type

100

75

50

25

0

Animal Type


CA TX NY FL IL PA OH MI GA NC NJ VA WA AZ MA IN TN MOMD WI MN CO AL SC LA KY OR OK CT IA MS AR KS UT NV NMWV NE ID ME NH HI RI MT DE SD AK ND VT DC WY0.0

5.0M

10M

15M

20M

25M

30M

35M

Popu

latio

n 65 Years and Over

45 to 64 Years

25 to 44 Years

18 to 24 Years

14 to 17 Years

5 to 13 Years

Under 5 Years

Stacked Bar Charts


[Bostock, 2012]

Stacked Bar Charts• Data: multidimensional table: one quantitative, two categorical • Task: lookup values, part-to-whole relationship, trends • How: line marks: position (both horizontal & vertical),

subcomponent line marks: length, color • Scalability: main axis (hundreds like bar chart), bar classes (<12)

• Orientation: vertical or horizontal (swap how horizontal and vertical position are used.


than 6,000 data sets at once. While the layout method of the Name-Voyager was not novel—it used a standard stacked graph layout, with some level-of-detail calculations—the popular response to the applet suggested that stacked graphs have the ability to engage mass audiences.

A follow-up design to the NameVoyager, described in [20], showed hierarchical time series. That is, it used interactivity and color to display time series that were arranged into categories and subcategories. In the Many Eyes system [17], this technique was made broadly available on the web.

A final related work is the Revisionist [7] visualization of changes in source code over time. While not technically a stacked graph, the geometry is related since each line of code is represented by a curved stripe. Revisionist minimizes visual distortion by having a curved baseline that allows the visualization to roughly align identical lines of code between releases.

3 LAST.FM AND THE NEW YORK TIMES

3.1 Listening History - Last.fm

Listening History was created by the first author for a class project at Carnegie Mellon University. The six-week assignment was to collect and display a data set in an interesting and novel way. As described in the introduction, Listening History [4] visualizes trends in an individual’s music listening, as derived from data in the last.fm service. The x-axis represents time and each stripe represents an artist. The thickness of a stripe shows the number of times that songs from the artist were listened to in a given week. The color, as detailed in section 5, encodes two dimensions: the saturation is determined by the overall number of times an artist is listened, and the hue is related to the earliest date at which one of the artist’s songs were heard.

A critical design goal for this visualization was to create a graphic that did not look scientific or mathematical, but rather felt organic and emotionally pleasing. In section 5 we will see that, ironically, achieving this goal relied on significant computation. A side effect of the algorithm is the signature asymmetry between the top and bottom curves which form the organic shape and, as discussed later, minimizes internal distortion.

At the end of the course, a few large-scale posters, some over 12 feet long, were printed as holiday gifts. The reaction of the recipients provides evidence, if anecdotal, that the graphic succeeded in elicit-ing strong emotional reactions when people saw their own listening history. People often remarked at the ability to see critical life events reflected in their music listening habits.

One pointed to the beginning and end of three separate relation-ships, and how his listening trends changed dramatically. Another noted the moment when her dog had died, and the resulting impact on the next month of listening. A third pointed out his dramatic differ-ences between summer and winter listening trends. As in the Themail system of Viégas et al. [18], the visualization of historical and per-sonal data seemed effective at eliciting reflective storytelling.

After Listening History was made public, there was high demand for personalized versions of these graphics by other last.fm members. In fact this demand was so strong that a number of imita-tors emerged, including Maya’s Extra Stats [12] and Godwin’s Last Graph [13] Interestingly, these services and other imitators use the simpler ThemeRiver layout and a simpler color scheme.

The popularity of these imitators (Last Graph has created visu-alizations for more than 24,000 users) suggests the hypothesis that stacked graphs have an ability to communicate large amounts of data to the general public in an intriguing and satisfactory way.

3.2 New York Times - Box Office Revenue

The Box Office Revenue graph, created by the first author and the graphics department of the Times [2,6] highlighted the dichotomy between box office hits and Oscar nominations, discussed in the orig-inal article. The printed graphic ran vertically to best use the avail-able space, time running top to bottom; the online version ran left to right. To allow a quick reading of the graph, coloring was much simpler than in Listening History: a discrete palette signified ranges of overall revenue. Furthermore, stroke lines were added because of issues with print registration.

The online response to these graphics was intense and rapid. Many blogs and social websites featured long lists of comments dis-cussing data-points shown in the graph. As with the NameVoyager, blog posters and their commenters engaged in a social style of data analysis and critique of the new visual form. What follows are anec-dotes discussing these visualizations, which provide a rough sense of the breadth and intensity of the online response.

Individual bloggers often found particular discoveries and pointed them out to their readers. For example, one said:

C1: note the double spike on ‘Harry Potter an the Order of the Phoenix’. And the long hump on ‘Alvin and the Chipmunks’. ‘Juno’ also has an interesting curve as it did almost nothing for a month before popping out later in it’s run. Though that may be because it was released in just enough theaters to become Oscars fig 1 – section from Listening History of primary author

fig 2 – films from the summer of 2007

Streamgraphs• Include a time attribute • Data: multidimensional table, one

quantitative attribute (count), one ordered key attribute (time), one categorical key attribute

• + derived attribute: layer ordering (quantitative)

• Task: analyze trends in time, find (maxmial) outliers

• How: derived position+geometry, length, color

• Scalability: more categories than stacked bar charts


[Byron and Wattenberg, 2012]

NYTimes “Ebb and Flow of Movies"


[http://www.nytimes.com/interactive/2008/02/23/movies/20080223_REVENUE_GRAPHIC.html]

http://www.nytimes.com/interactive/2013/05/20/us/oklahoma-tornado-map.html

Dot and Line Charts• Data: one quantitative attribute, one

ordered attribute • Task: lookup values, find outliers

and trends • How: point mark and positions

• Line Charts: add connection mark (line)

• Similar to scatterplots but allow ordered attribute


20

15

10

5

0

Year20

15

10

5

0

Year[Munzner (ill. Maguire), 2014]

10-year-olds 12-year-olds

60

50

40

30

20

10

0

60

50

40

30

20

10

0 10-year-olds 12-year-olds

Proper Use of Line and Bar Charts


[Zacks and Tversky, 1999, Munzner (ill. Maguire), 2014]

Proper Use of Line and Bar Charts


Female Male

60

50

40

30

20

10

0 Female Male

60

50

40

30

20

10

0

[Zacks and Tversky, 1999, Munzner (ill. Maguire), 2014]

Aspect Ratio• Trends in line charts are more apparent because we are using angle

as a channel • Perception of angle (and the relative difference between angles) is

important • Initial experiments found people judge differences in slope when

angles are around 45 degrees (Cleveland et al., 1988, 1993)


• Use frequency analysis to determine aspect ratios • Add trend curves using a low-pass filter

Multiscale BankingIEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 12, NO. 5, SEPTEMBER/OCTOBER 2006

Sunspot Cycles Aspect Ratio = 3.96

Aspect Ratio = 22.35

Power Spectrum

Aspect Ratios

Figure 5. Sunspot observations, 1700-1987. The first plot shows low-frequency oscillations in the maximum values of sunspot cycles. The second plot brings the individual cycles into greater relief. Carbon Dioxide Measurements Aspect Ratio = 1.17

Aspect Ratio = 7.87

Power Spectrum

Aspect Ratios

Figure 6. Monthly atmospheric CO2 measurements. The first plot shows a baseline trend of increasing values, with a slight inflection. The second plot more clearly communicates the yearly oscillations. Figures 5-8 show the results of applying multi-scale banking to real-world data sets. Data sets are plotted at each computed aspect ratio, with banked trend lines shown in red. The power spectrum plot shows a frequency-domain representation of the data, annotated with potential scales of interest. The aspect ratio plot shows the banked aspect ratios for each possible lowpass filtering of the data, annotated with the final aspect ratios returned by the algorithm.

PRMTX Mutual FundAspect Ratio = 4.23


Power Spectrum

Aspect Ratios

Figure 7. PRMTX mutual fund performance, 1997-2006. The first plot shows the boom and bust of the “dot-com” bubble and subsequent recovery. Tthe second plot affords closer consideration of short-term variations. Downloads of the prefuse toolkit Aspect Ratio = 1.44

Aspect Ratio = 2.89

Aspect Ratio = 8.81

Power Spectrum

Aspect Ratios

Figure 8. Daily download counts of the prefuse visualization toolkit. The first plot shows a general increase in downloads. The second plot shows weekly variations, including reduced downloads on the weekends. The third plot enables closer inspection of day-to-day spikes and decays.

706


[Heer and Agrawala, 2006]

IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 12, NO. 5, SEPTEMBER/OCTOBER 2006

Sunspot Cycles Aspect Ratio = 3.96


Power Spectrum

Aspect Ratios

Figure 5. Sunspot observations, 1700-1987. The first plot shows low-frequency oscillations in the maximum values of sunspot cycles. The second plot brings the individual cycles into greater relief. Carbon Dioxide Measurements Aspect Ratio = 1.17

Aspect Ratio = 7.87

Power Spectrum

Aspect Ratios

Figure 6. Monthly atmospheric CO2 measurements. The first plot shows a baseline trend of increasing values, with a slight inflection. The second plot more clearly communicates the yearly oscillations. Figures 5-8 show the results of applying multi-scale banking to real-world data sets. Data sets are plotted at each computed aspect ratio, with banked trend lines shown in red. The power spectrum plot shows a frequency-domain representation of the data, annotated with potential scales of interest. The aspect ratio plot shows the banked aspect ratios for each possible lowpass filtering of the data, annotated with the final aspect ratios returned by the algorithm.

PRMTX Mutual FundAspect Ratio = 4.23


Power Spectrum

Aspect Ratios

Figure 7. PRMTX mutual fund performance, 1997-2006. The first plot shows the boom and bust of the “dot-com” bubble and subsequent recovery. Tthe second plot affords closer consideration of short-term variations. Downloads of the prefuse toolkit Aspect Ratio = 1.44

Aspect Ratio = 2.89

Aspect Ratio = 8.81

Power Spectrum

Aspect Ratios

Figure 8. Daily download counts of the prefuse visualization toolkit. The first plot shows a general increase in downloads. The second plot shows weekly variations, including reduced downloads on the weekends. The third plot enables closer inspection of day-to-day spikes and decays.

706 Multiscale Banking


[Heer and Agrawala, 2006]

As far as we can tell, there is no other work in perception on sloperatio estimation. However, there is substantial work on related percep-tual estimation problems that might provide insight. Fisher [9] foundthat people under and over estimated the orientation of a single 2D lineby as much as 3 degrees. The effect was largest for orientations near45°. Kennedy et al. [13] showed that angle perception is biased by thelengths of the angle’s legs. The effect was as large as 15 degrees forextremely scalene angles. Recent work on 2D angle perception haveexplained the traditional effect that small angles are overestimated andbig angles are underestimated as a result of a cognitive process that at-tempts to perceive the angles in 3D instead of 2D. The distribution ofangles that occur in the real-world bias this cognitive process to under-or overestimate 2D angles [14, 12]. The effect is up to 2 degrees and islargest for angles near 20°. Fermuller and Malm proposed a unifyingimage processing-based explanation for many types of illusions [7].

2.2 Aspect Ratio Algorithms

Banking to 45° has led to the design of automated algorithms for pick-ing good aspect ratios. Since it is not clear how to extend the Cleve-land et al. experimental results from plots with two line segments toplots with many line segments, multiple algorithms have been recom-mended. In the original paper, it was suggested that the aspect ratioof plots should be chosen such that the median segment slope was 1,an approach they called median absolute slope (MS). Cleveland latersuggested an alternative method, length-weighted average orientation(AWO), that sets the length-weighted average of the absolute segmentorientations (the angle made with the horizontal) to 45° [3]. Later,Heer and Agrawala [10] proposed the global orientation resolution(GOR) method that selects the aspect ratio by maximizing the sum ofsquares of the angles between all pairs of segments in the plot. A com-putationally cheaper approach is to only consider the angles betweenadjacent pairs of segments. Heer and Agrawala called this approachthe local orientation resolution (LOR). Recently, Talbot et al. [15] rec-ommended an approach based on minimizing the arc length of the datacurve and demonstrated that this approach is more robust than previ-ous work.

2.3 Line Chart Visualization

Beattie and Jones [1] examine business graphics in light of the sloperatio hypothesis and conclude that viewer perception of business per-formance is more accurate when graphs are banked to 45°. Best [2]compare how extrapolation of visual trends varied across a number ofdifferent chart types including line charts and find that the error de-pended upon the trend’s functional form. Correll et al. [6] explorehow aggregate comparisons are made in line charts and suggest analternate visualization based on a color encoding to make these com-parisons easier. Horizon plots [8, 11] permit compressing the verticalaxis of a line chart without changing the aspect ratio by overlayinghorizontal slices of the plot.

3 RESEARCH GOALS AND METHODS

Our objective is to better understand the theory behind aspect ratioselection. To do this, we revisit the question raised in the originalCleveland et al. study: how does slope ratio estimation accuracy de-pend on the true slope ratio and on the aspect ratio of the line segments(expressed as the mid-angle between the segments)? Our approach toanswering this question is twofold:

1. Based on insights gained from a series of pilot studies (Sec-tion 4), we develop an empirical model of human slope ratiojudgments (Section 5).

2. Then, through two formal experiments, we demonstrate that themodel fits observed data well and we provide a perceptual inter-pretation of its components (Sections 6 and 7).

In our model building process and in our experiments we follow thehigh-level design of the original study. However, we modify the studydesign to address two major limitations of their work.

11%

17%

25%

35%

48%

65%

87%

9° 18° 30° 45° 60° 72° 81°

Mid−angle (θm)

Slo

pe r

atio

(p

ij)

Fig. 2. Space of line comparisons parameterized by mid-angle and anon-linear transformation of the true slope ratio. Line pairs used in thisstudy are plotted here. The spacing on the x-axis is not linear due to achange in our preferred parameterization during post-experiment dataanalysis. Line pairs used in Cleveland et al. are indicated with bluepoints.

First, the original design only covers a small portion of the spaceof slope pairs. In our work, we sample a much larger portion of thespace. The range of line pairs studied in our experiment is shown inFigure 2 parameterized by the mid-angle, θm, the angle halfway be-tween the line segments, and by the slope ratio (expressed as a per-centage), pi j, between two segments i and j. (Due to a change inparameterization part way through our analysis, our mid-angle sam-ples are spaced slightly unevenly.) In comparison, the set of line pairsstudied in Cleveland et al., shown with blue points in Figure 2, spans amuch smaller range. To keep the sample count manageable, we sampleless frequently along the slope ratio-axis than Cleveland et al; however,our pilot studies and Figure 6 in Cleveland and McGill [5] indicate thatthe slope estimation error function changes slowly in this direction soa lower sampling rate makes better use of limited subject time.

Second, in the Cleveland et al. design, subjects were explicitly in-structed to make their slope estimates by comparing the heights (y-extent) of both lines. This approach is exactly correct if the x-extent ofboth lines is the same. However, we wonder if this is how most visu-alization users would naturally approach the slope estimation task. Toexplore how users make slope comparisons in practice, we omit thisinstruction in our studies, instead allowing subjects to use their ownapproach to the problem.

4 PILOT STUDIES

To help us better understand the slope ratio estimation task, we ran aseries of 11 pilot studies on Amazon Turk. We used these informal,rapid studies to quickly iterate on our experimental design and to ex-plore a variety of model possibilities.

For example, in early iterations we used a study design with 100distinct slope comparison tasks (the product of 10 mid-angles and 10slope ratios), but found that this resolution was unnecessary since theshape of the response function was relatively smooth. In later itera-tions, we settled on the design shown in Figure 2 with 49 (7×7) slopecomparisons, allowing us to run double the number of replications atthe same cost. We also found that linear sampling of the slope ratioresulted in many line pairs that were visually indistinguishable. So wemodified our design to vary the true slope ratio nonlinearly making thechange in the visual angle between line segments equal.

Expanding the Study• Cleveland et al. did not study the

entire space of slope comparisons and 45 degrees was at the low end of their study (blue marks on right)

• Talbot et al. compared more slopes and found that people do better with smaller slopes

• Baselines may aid with this


[Talbot et al., 2013]

Heatmaps• Data: Two keys, one quantitative

attribute • Task: Find clusters, outliers,

summarize • How: area marks in grid, color

encoding of quantitative attribute • Scalability: number of pixels for area

marks (millions), <12 colors • Red-green color scales often used

- Be aware of colorblindness!


[fastpitchanalytics.com]

Fast-Pitch Softball Slugging Percentage

http://fastpitchanalytics.com

Bertin Matrices• Must we only use color?

- What other marks might be appropriate?


[C.Perrin et al., 2014]

Fig. 6. Using a crosset to adjust the range of three rows at a time.

Figure 6). Compared to standard spreadsheet interactions for resizingrows or columns, the resizing crosset does not require the selectionto be specified in advance. Since this operation impacts the layout ofcrossets, the tool only shows a partial preview. The same is true forseparator adjustment tools and the automatic reordering tool.

By supporting crossing-based interactions, BERTIFIER makes it pos-sible to quickly change arbitrary groups of adjacent rows and columns,which is useful in many cases, including for automatically reorder-ing rows or columns within identified groups. Crossing all rows orcolumns allows users to apply operations to the entire table in a quasi-instantaneous manner, provided the table fits the screen. All theseinteractions remain consistent across all tools.

Crossets extend previous work on crossing-based interfaces in sev-eral ways [1, 2, 5]. One problem with such interfaces is that they requiresteering, a slow motor task [1]. Crossets do not have this problem sincethe command is selected on mouse press, after which the user is allowedto freely deviate from a straight trajectory. As far as we know, crossinggestures have never been applied to manipulate multiple sliders at once,and have never been used to interact with tables.

4.3 Human-Assisted ReorderingHuman-Assisted reordering is supported both through manual reorder-ing interactions and through automatic visual reordering.

4.3.1 Manual Reordering Interactions

As in previous implementations [14, 45, 46, 49], we support columnand row reordering by drag and drop. This is the first level of integra-tion between automatic and manual reordering, and allows to tweakthe results of automatic reordering [33]. Following previous findingsthat reordering rows and columns concurrently can be confusing tousers [49], we lock the reordering on the row or column based onthe initial dragging direction. To help users understand changes, weperform animated transitions during the dragging operation.

Following previous recommendations [49–51], we also support drag-ging on sets of rows and columns. We provide a tool that lets users“glue” several rows or columns together (Figure 5 bottom right). Sincewe distinguish between groups used for concurrent manipulation andvisual groups (i.e., separators), we avoided the ambiguous term “group”.Our automatic reordering algorithm preserves rows and columns gluedtogether by the user, thus providing a second level of integration.

4.3.2 Automatic Visual Reordering

The principle of automatic visual reodering is that rows and columnsare reordered not according to the underlying data, but according totheir visual similarity. We believe this principle is easier to understandfor users not familiar with data analysis. Visual reordering is only auser interface metaphor that does not have to accurately capture whatthe system is doing, but which is meant to elicit a simple and “goodenough” mental model of the system to allow for easy tuning.

The implementation of this metaphor is fairly simple and relies ontwo basic principles: i) the reordering algorithm should take as inputthe data after it has been conditioned and normalized (e.g., betwewn0 and 1), and ii) the visual encodings used should ensure that visualdifferences are roughly proportional to numerical differences. Fromthis it follows that the automatic reordering algorithm will behave as ifit were operating visually. Next we discuss to what extent the conditionii) is fulfilled by the visual encodings implemented in BERTIFIER.

0 1 2 3 4 5 6 7 8 9 10 11 textText

Grayscale

Circle

Line

Bar chart

Black and whitebar chartAverage

bar chart

Dualbar chart

QUANTITYOF INK

ENCODINGS

POSITIONALENCODING

MEAN-BASEDENCODINGS

Fig. 7. Encodings in BERTIFIER. The range of each row is set to [1, 10]. 0and 11 are beyond the range. Crosses encode N/A values.

4.3.3 Visual Encodings

BERTIFIER implements eight types of visual encodings (Figure 7). Forthe sake of generality, we consider text as being one type of encoding.Automatic reordering is however disabled on text to reinforce the visualreordering metaphor. All encodings except grayscale have been eitherexplicitly mentioned in Bertin’s book [10] or were used by him. Thesoftware can be easily extended with other types of encodings.

The black and white bar charts and average bar charts are particularin that they encode summary statistics in addition to value: valueslower than the mean are shown in white or gray, while values above themean are shown in black. The line encoding is also different in that ituses a positional encoding. These three encodings have been only usedoccasionally and are not implemented in any of the physical matrices.

The remaining encodings are more common and all roughly followthe rule that the quantity of ink—the average pixel darkness—is propor-tional to the normalized data value. We enforced this rule to make theencodings consistent with the visual reordering metaphor. For example,the dual bar chart encodes the values from white to black, with a fullyhatched cell corresponding to the mean value of the row. In this case,we use a 50% hatching to respect the rule of proportionality. Thus forthese four encodings, computing a numerical difference between cellsis equivalent to computing the difference in their average shading. Therule of proportionality is challenging to implement for the “circle” en-coding, since the circle is clipped and progressively turns into a square.We chose to replicate Bertin’s original encoding although previouswork described an optimal scale for symbol size discrimination [38].Deriving the correct circle radius is a geometrical problem without anyanalytic solution, but we found the following good approximation:

r =

(D = 2

pv/⇡ if v ⇡

412 (t+ t6)(

p2� 1) + 1 with t = D�1

2/⇡�1 if v > ⇡4

with v being the cell value and r the (unclipped) circle radius.

4.3.4 Interactive Data Conditioning

Although users can pre-process their data in their spreadsheet as partof step S1, BERTIFIER provides tools for performing further dataconditioning of rows on-the-fly (the “Adjust” toolbar in Figure 5).Slider crossets are provided for i) adjusting the data range and clippingthe values accordingly, ii) reducing the maximum value of the data(which has the effect of making all cells look brighter or disappear),iii) turning values into discrete steps. In addition, users can iv) invertrow or columns values. i), iii) and iv) have been recommended byBertin [12] and i), ii) and iii) are implemented in CHART [7].

These operations change the matrix visually in a similar way asphoto retouching tools. Since the automatic reordering is performedon post-processed values, these controls provide a way of fine-tuningthe reordering algorithm without explicitly tuning any of its internalparameters. Specifically, rows that are made brighter will be given alower weight by the reordering algorithm. Rows that are made entirelywhite will be ignored. A specific case of this is making all rows whiteexcept one, which enables a regular sorting operation.

Sometimes it is useful to provide custom ranges (e. g., for applying auniform range to all rows). To achieve this, users can specify the rangeof each row in the initial spreadsheet, using special header names. Acrosset in BERTIFIER allows to enable or disable this custom range.

Bertin’s Encodings


[C.Perrin et al., 2014]

Matrix Reordering


[Bertin Exhibit (INRIA, Vis 2014), Photo by Robert Kosara]

https://eagereyes.org/blog/2014/vis-2014-wednesday

Cluster Heatmap


[File System Similarity, R. Musăloiu-E., 2009]

http://cs.jhu.edu/~razvanm/fs-expedition/

Cluster Heatmap• Data & Task: Same as Heatmap • How: Area marks but matrix is ordered by cluster hierarchies • Scalability: limited by the cluster dendrogram

• Dendrogram: a visual encoding of tree data with leaves aligned


Sepal.Length

2.0 3.0 4.0

●●

●●

●

●

●

●

●

●

●

●●

●

● ●●

●

●

●●

●

●

●●

● ●●●

●●

●●●

●●

●

●

●

●●

● ●

● ●●

●

●

●●

●

●

●

●

●

●

●

●

●

●●

●● ●

●

●

●●

●

●●

●●

●●●

● ●

●●

●●●●

●

●

●

●

●●●

●●

●

● ●●

●

●

●

●

●

●

●●

●

●

●

●

●

●●

●

● ●

●●

●●

●

●

●

●

●

●

●

● ●●

●●

●

●●●

●

●●

●

●●●

●

●●●

●●

●●

●●●●

●

●

●

●

●

●

●

●●

●

●●●●

●

●●

●

●

●●

●●●●

●●

●●●

●●

●

●

●

●●

●●

●●●●

●

●●

●

●

●

●

●

●

●

●

●

●●

●●●

●

●

●●

●

●●

●●

●●●

●●

●●●●●

●

●

●

●

●

●●●

●●

●

●●●

●

●

●

●

●

●

●●

●

●

●

●

●

●●

●

●●

●●

●●

●

●

●

●

●

●

●

●●●

●●

●

●●●

●

●●

●

●●

●

●

●●●

●●●

●

0.5 1.5 2.5

4.5

5.5

6.5

7.5

●●●●

●

●

●

●

●

●

●

●●

●

● ●●

●

●

●●

●

●

●●● ●●●

●●

●●●

●●

●

●

●

●●

●●

●●●

●

●

●●

●

●

●

●

●

●

●

●

●

●●

●● ●

●

●

●●

●

●●

●●

●●●● ●

●●●●●

●

●

●

●

●

●●●

●●

●

●●●

●

●

●

●

●

●

●●

●

●

●

●

●

●●

●

● ●

●●

●●

●

●

●

●

●

●

●

●●●

●●

●

●●●

●

●●

●

●●

●

●

● ●●

●●

●●

2.0

3.0

4.0

●

●●●

●

●

● ●

●●

●

●

●●

●

●

●

●

●●

●

●●

●●

●

●●●

●●

●

●●

●●

●●

●

●●

●

●

●

●

●

●

●

●

●●●●

●

●●

●

●

●●

●

●

●

●●●

●

●

●

●

●

●

●

●●●●●

●

●●●

●●

●

●

●

●

●

●●

●

●

●

●

●● ●

●

●

●

●

●●● ●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

● ●●

●●

●●

●●●

●

●●●

●

●

●●

●●●

●

●●

●

●

●

●

●Sepal.Width

●

●●●

●

●

●●

●●

●

●

●●

●

●

●

●

●●

●

●●

●●

●

●●●●●

●

●●

●●

●●

●

●●

●

●

●

●

●

●

●

●

●●●●

●

●●

●

●

●●

●

●

●

●●●●

●

●

●

●

●

●

●●●

●●

●

●●●

● ●

●

●

●

●

●

●●

●

●

●

●

●●●

●

●

●

●

●●● ●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

● ●●

●●

●●

●●●

●

●●●

●

●

●●

●●●

●

●●

●

●

●

●

●

●

●●●

●

●

●●

●●

●

●

●●

●

●

●

●

●●

●

●●

●●

●

●●●●●

●

●●

●●

●●

●

●●

●

●

●

●

●

●

●

●

●●●●

●

●●

●

●

●●

●

●

●

●●●●

●

●

●

●

●

●

●●●●

●●

●●●

● ●

●

●

●

●

●

●●

●

●

●

●

●●●

●

●

●

●

●●

●●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●●●

●●

●●

●●

●

●

●●●

●

●

●●

● ●●

●

●●

●

●

●

●

●

●●●● ●●

● ●● ● ●●●● ●

●●●●● ●●

●

●●●●●●●● ●● ●●● ●●● ●●●●●●

● ●● ●●

●●●

●

●● ●

●

●

●●

●●

●

●

●●●

●

●

●

●

●●●●

●●●

●●● ●

●

● ● ●●

●●● ●

●

●

●●● ●

●

●

●

●

●●●

●

●

●●

●

●●●

●●●●

●●

●

●

●

●

●

●●

●●

● ●●

●

●●

●●

●●

●

●●●●

●●●●●●

●

●● ●● ●●

●●● ● ●●●● ●

●●●●●● ●

●

●●● ●●●●● ● ●●●● ●●● ●●● ●

●●

● ●● ●●

●●●

●

●● ●

●

●

●●

●●

●

●

●●●

●

●

●

●

● ●●●

● ●●

●●● ●

●

● ●●●

●●● ●

●

●

● ●●●

●

●

●

●

●●●

●

●

●●

●

●● ●● ● ●●

●●

●

●

●

●

●

●●

● ●

● ●●

●

●●

●●

●●

●

●●●●

●●●● ● ●●

Petal.Length

12

34

56

7

●●●●●●

●●●●●●●●●

●●●●●● ●

●

●●● ●●●●● ●●●●●●●●●●●●

●●

●●●●●

●●●

●

●● ●

●

●

●●

●●

●

●

●●●

●

●

●

●

●●●●● ●●

●●● ●

●

●●●●●●

● ●

●

●

●●●●

●

●

●

●

●● ●

●

●

●●

●

●●●

● ●●●

●●

●

●

●

●

●

●●

●●

●●●●

●●

●●●●

●

● ●●●

● ●●●●●

●

4.5 5.5 6.5 7.5

0.5

1.5

2.5

●●●● ●●

●●●●

●●●●

●●●

● ●●●

●●

●

●●●●●●●●

●●●● ●

●● ●

●●●

●●

●●● ●●

●● ●

●●

●

●

●

●●

●

●

●

●●

●●

●

●

●

●

●●

●●●●

●●

●●●

●

●●

●●

●●●●

●●

●

●●● ●

●●

●

●●

●

●●

●●●

●

●●

●●

●●

●

●●

●

●

● ●●

●

●●●

●

●

●●

●

●●

●●

●●

●

●●

●

●●●

●●

●

●

●● ●● ●●

●●●

●●●

●●●

●●● ●●

●●

●

●

●●●●●●●

●

●●●● ●

●● ●

●●●

●●

●●● ●●

●●●

●●●

●

●

●●

●

●

●

●●

●●

●

●

●

●

●●

●●●●

●●

●●●

●

●●

●●

● ●●●

●●

●

●●

●●●

●

●

●●

●

●●

●●●

●

●●

●●

●●

●

●●

●

●

●●●

●

●● ●

●

●

●●

●

●●

●●

●●

●

●●

●

●●

●

●●

●

●

1 2 3 4 5 6 7

●●●●●●

●●●●●●●●

●●●●●●●

●●

●

●●●●●●●●

●●●●●●●●●●●

●●

●●●●●

●●●

●●●

●

●

●●

●

●

●

●●

●●

●

●

●

●

●●

●●●●

●●

●●●●

●●●●

●●●●●

●●

●●●●

●●

●

●●

●

●●

●●●

●

●●●

●

●●

●

●●

●

●

● ●●

●

●●●

●

●

●●

●

●●

●●

●●

●

●●

●

●●

●

●●

●

●

Petal.Width

Iris Data (red=setosa,green=versicolor,blue=virginica)

Scatterplot Matrix (SPLOM)• Data: Many quantitative attributes • Derived Data: names of attributes • Task: Find correlations, trends,

outliers • How: Scatterplots in matrix

alignment • Scale: attributes: ~12, items:

hundreds?

• Visualizations in a visualization: at high level, marks are themselves visualizations…


[R command from Wikipedia]

http://commons.wikimedia.org/wiki/File:Anderson%27s_Iris_data_set.png#Creation

Spatial Axis Orientation• So far, we have seen the vertical

and horizontal axes (a rectilinear layout) used to encode almost everything

• What other possibilities are there for axes?



Math Physics Dance Drama

1009080706050 40302010

0

Parallel Coordinates



• What other possibilities are there for axes?- Parallel axes




1009080706050 40302010

0

Parallel Coordinates



• What other possibilities are there for axes?- Parallel axes- Radial axes



Parallel Coordinates• Data: many quantitative attributes • Task: Find trends, extremes, correlation • How: vertical spatial position for each attribute, connection marks

for identity, axes horizontally spaced • Scalability: <40 attributes, hundreds of values

• Connection marks help visualize trends between particular values • Ordering the horizontal axes is important • Not as well-known, often requires learning


Math

Physics

Dance

Drama



1009080706050 40302010

0

Scatterplot Matrix Parallel Coordinates

Comparing SPLOMs and Parallel Coordinates



Correlation in Parallel Coordinates


[Wegman, 1990]

1.0

0.8

0.2

0

-0.2

-0.8

-1

Overdraw in Parallel Coordinates


dense areas appear emphasized. The disadvantage of this method isthat it relies on overlapping points/lines to identify clusters. Clus-ters without overlapping elements will not be visually emphasized.Keim et al. [13] studied pixel-level visualization schemes which

permit the display of a large number of records on a typical worksta-tion screen based on recursive layout patterns. However, the num-ber of displayable records is dependent on the size of the displayarea. This limitation restricts the scalability of their method. More-over, since each pixel only represents one variable, it is difficult toconvey the interactions among variables.Wills [30] describes a visualization technique for hierarchical

clusters. His approach expands upon the tree-map idea [24] byrecursively subdividing the tree based on a dissimilarity measure.However, the main purpose is to display the clustering results, andin particular, the data partitions at a given dissimilarity value.Our research draws on several of the ideas found in the above

work. As in [31], we store and present our data at multiple res-olutions. However, to overcome the data ordering limitation ofwavelets, we use clustering and partitioning techniques. We alsouse the opacity of lines as in [28] to reduce clutter. However, ratherthan conveying data density with overlapping lines, we use data ag-gregation techniques to collapse data into clusters, and show thepopulation and extents of clusters with bands of varying translu-cency.

3 Parallel Coordinates

Figure 1: Parallel coordinates of Detroit homicide data set: a 7-dimensional data set with 13 records. Notice that there are inversecorrelations between the number of cleared homicides and both thenumber of government workers and the total number of homicides.

We have chosen parallel coordinates as the visualization tech-nique to extend upon to support large scale data. Parallel coordi-nates is a technique pioneered in the 1980’s which has been appliedto a diverse set of multidimensional problems [10, 27]. It has sincebeen incorporated into many commercial and public-domain sys-tems, such as WinViz [17], XmdvTool [25, 19], and SPSSDiamond(http://www.spss.com/software/diamond).In this technique, each data dimension is represented as a (hori-

zontal or) vertical axis, and the axes are organized as uniformlyspaced lines. A data element in an -dimensional space is mappedto a polyline that traverses across all of the axes crossing each axisat a position proportional to its value for that dimension.Parallel coordinates have a distinct advantage over conventional

orthogonal coordinates. By laying out the vertical axes horizontallyacross the screen, the number of dimensions that can be visualizedis restricted only by the horizontal resolution of the screen. This isin contrast to multivariate visualization in orthogonal coordinateswhere previous work [21] has attempted to augment each spatialpoint with a vector of values, usually with some visual icon that en-codes the values. It is clear that with such an encoding scheme, one

soon runs out of encoding possibilities as the number of dimensionsincreases. The issue of dimensionality never arises in parallel coor-dinates, though as the axes get closer it may become more difficultto perceive structures or data relations. Moreover, using parallel co-ordinates, we can easily spot correlations between variables in thedata set (see Figure 1).The main difficulty of directly applying parallel coordinates to

large data sets is that the level of clutter present in the visualizationreduces the amount of useful information one can perceive. Forexample, the display of a mass of overlapping lines precludes theperception of relative densities present in the data set (see Figure2). Our approach reduces the amount of clutter by imposing a hier-

Figure 2: Parallel coordinates display of a Remote Sensing data set:a 5-dimensional data set with 16,384 records. Note the amount ofover-plotting precludes the perception of any data trends, correla-tions or anomalies.

archical organization on the data set. We then display aggregationsof the data at different levels of abstraction and provide tools fordynamically navigating and filtering the hierarchy, as described indetail in the following sections.

4 Overview of Hierarchical Parallel Coor-dinates

Exploratory data analysis is the summarization, display and manip-ulation of data to make it more comprehensible to human minds,thus uncovering underlying structure in the data and detecting im-portant departures from that structure [3]. A complete data ex-ploratory system thus has three major ingredients. Table 1 lists thethree major components to our approach and their correspondingsub-components. The three basic components are: the summariza-

Summarization Display Manipulation/Filtering

Hierarchical organizationwith statistical aggrega-tion

Proximity-Based ColoringTranslucency

Structure-based BrushDrill-Down/Roll-UpDimension ZoomingExtent ScalingDynamic Masking

Table 1: Basic Components of the proposed Hierarchical ParallelCoordinates Approach

tion of the data by imposing a hierarchical structure on the data set,a scheme for displaying -dimensional aggregate information anda set of tools for navigating, manipulating and filtering the hierar-chical structure. We shall describe each of these components in thefollowing sections.

dense areas appear emphasized. The disadvantage of this method isthat it relies on overlapping points/lines to identify clusters. Clus-ters without overlapping elements will not be visually emphasized.Keim et al. [13] studied pixel-level visualization schemes which

permit the display of a large number of records on a typical worksta-tion screen based on recursive layout patterns. However, the num-ber of displayable records is dependent on the size of the displayarea. This limitation restricts the scalability of their method. More-over, since each pixel only represents one variable, it is difficult toconvey the interactions among variables.Wills [30] describes a visualization technique for hierarchical

clusters. His approach expands upon the tree-map idea [24] byrecursively subdividing the tree based on a dissimilarity measure.However, the main purpose is to display the clustering results, andin particular, the data partitions at a given dissimilarity value.Our research draws on several of the ideas found in the above

work. As in [31], we store and present our data at multiple res-olutions. However, to overcome the data ordering limitation ofwavelets, we use clustering and partitioning techniques. We alsouse the opacity of lines as in [28] to reduce clutter. However, ratherthan conveying data density with overlapping lines, we use data ag-gregation techniques to collapse data into clusters, and show thepopulation and extents of clusters with bands of varying translu-cency.

3 Parallel Coordinates

Figure 1: Parallel coordinates of Detroit homicide data set: a 7-dimensional data set with 13 records. Notice that there are inversecorrelations between the number of cleared homicides and both thenumber of government workers and the total number of homicides.

We have chosen parallel coordinates as the visualization tech-nique to extend upon to support large scale data. Parallel coordi-nates is a technique pioneered in the 1980’s which has been appliedto a diverse set of multidimensional problems [10, 27]. It has sincebeen incorporated into many commercial and public-domain sys-tems, such as WinViz [17], XmdvTool [25, 19], and SPSSDiamond(http://www.spss.com/software/diamond).In this technique, each data dimension is represented as a (hori-

zontal or) vertical axis, and the axes are organized as uniformlyspaced lines. A data element in an -dimensional space is mappedto a polyline that traverses across all of the axes crossing each axisat a position proportional to its value for that dimension.Parallel coordinates have a distinct advantage over conventional

orthogonal coordinates. By laying out the vertical axes horizontallyacross the screen, the number of dimensions that can be visualizedis restricted only by the horizontal resolution of the screen. This isin contrast to multivariate visualization in orthogonal coordinateswhere previous work [21] has attempted to augment each spatialpoint with a vector of values, usually with some visual icon that en-codes the values. It is clear that with such an encoding scheme, one

soon runs out of encoding possibilities as the number of dimensionsincreases. The issue of dimensionality never arises in parallel coor-dinates, though as the axes get closer it may become more difficultto perceive structures or data relations. Moreover, using parallel co-ordinates, we can easily spot correlations between variables in thedata set (see Figure 1).The main difficulty of directly applying parallel coordinates to

large data sets is that the level of clutter present in the visualizationreduces the amount of useful information one can perceive. Forexample, the display of a mass of overlapping lines precludes theperception of relative densities present in the data set (see Figure2). Our approach reduces the amount of clutter by imposing a hier-

Figure 2: Parallel coordinates display of a Remote Sensing data set:a 5-dimensional data set with 16,384 records. Note the amount ofover-plotting precludes the perception of any data trends, correla-tions or anomalies.

archical organization on the data set. We then display aggregationsof the data at different levels of abstraction and provide tools fordynamically navigating and filtering the hierarchy, as described indetail in the following sections.

4 Overview of Hierarchical Parallel Coor-dinates

Exploratory data analysis is the summarization, display and manip-ulation of data to make it more comprehensible to human minds,thus uncovering underlying structure in the data and detecting im-portant departures from that structure [3]. A complete data ex-ploratory system thus has three major ingredients. Table 1 lists thethree major components to our approach and their correspondingsub-components. The three basic components are: the summariza-

Summarization Display Manipulation/Filtering

Hierarchical organizationwith statistical aggrega-tion

Proximity-Based ColoringTranslucency

Structure-based BrushDrill-Down/Roll-UpDimension ZoomingExtent ScalingDynamic Masking

Table 1: Basic Components of the proposed Hierarchical ParallelCoordinates Approach

tion of the data by imposing a hierarchical structure on the data set,a scheme for displaying -dimensional aggregate information anda set of tools for navigating, manipulating and filtering the hierar-chical structure. We shall describe each of these components in thefollowing sections.

[Fua et al., 1999]

Figure 4: This image sequence shows a Fatal Accident data set of 230,000 data elements at different level of details. The first image shows acut across the root node. The last image shows the cut chaining all the leaf nodes. The rest of the images show intermediate cuts at varyinglevels-of-detail. (See Color Plates).

Figure 6: Left image shows Iris data set without proximity-based coloring. Right image shows Iris data set with proximity-based coloringrevealing three distinct clusters depicted by concentrations of blue, green and pink lines. (See Color Plates).

Hierarchical Parallel Coordinates


[Fua et al., 1999]

cis 467/602-01: data visualizationdkoop/cis467/lectures/lecture11.pdf · simpler themeriver layout...

Documents