Data Visualization (CIS 468)
Multiple Views
Dr. David Koop
D. Koop, CIS 468, Fall 2018
Multiple Views
�2
[Improvise, Weaver, 2004]D. Koop, CIS 468, Fall 2018
Multiple Views• Why have just one visualization? • Sometimes data is best examined in more than one view
- Clutter/visual overload - Different attributes (cannot show all attributes in one view) - Different scales (task requires overview or detail) - Different encodings (no single encoding is optimal for all tasks)
• Eyes Beat Memory (Ch. 6) - Aiding working memory:
side-by-side/layers > animated > jump cuts - Showing all visual elements at once → don't need to remember
�3D. Koop, CIS 468, Fall 2018
Summary of Choices (Scatterplot + Bar Chart)
�4
[W. Javed and N. Elmqvist, 2012]D. Koop, CIS 468, Fall 2018
Technique Visualization A Visualization B Spatial Relation Data Relation
ComVis [24] (Figure 2) any any juxtapose noneImprovise [39] (Figure 3) any any juxtapose noneJigsaw [36] any any juxtapose noneSnap-Together [30] any any juxtapose nonesemantic substrates [34] (Figure 4) node-link node-link juxtapose item-itemVisLink [11] (Figure 5) radial graph node-link juxtapose item-itemNapoleon’s March on Moscow [37] time line view area visualization juxtapose item-itemMapgets [38] (Figure 6) map text superimpose item-itemGeoSpace [22] (Figure 7) map bar graph superimpose item-item3D GIS [8] map glyphs superimpose item-itemScatter Plots in Parallel Coordinates [45] (Figure 8) parallel coordinate scatterplot overload item-dimensionGraph links on treemaps [14] (Figure 9) treemap node-link overload item-itemSparkClouds [21] tag cloud line graph overload item-itemZAME [13] (Figure 10) matrix glyphs nested item-groupNodeTrix [17] (Figure 11) node-link matrix nested item-groupTimeMatrix [44] matrix glyphs nested item-groupGPUVis [25] Scatterplot glyphs nested item-group
Table 1: Classification of common composite visualization techniques using our design space.
(a) Juxtaposed views. (b) Integrated views. (c) Superimposed views. (d) Overloaded views. (e) Nested views.
Figure 12: Example of composing a scatterplot and bar graph using different methods.
datasets in the same space and using different visualizations, butalso highlights the relational linking between the two datasets.
Nested views provide an efficient approach to link each of thedata values, visualized through the host visualization, to its relateddataset, visualized through client visualizations. This is achievedby nesting clients inside the visual marks in the host.
• Benefits: Very compact representation, easy correlation.• Drawbacks: Limited space for the client visualizations, clut-
ter is high, and visual design dependencies are high.• Applications: Again, situations that call for augmenting a
particular visual representation with additional mapping.
Figure 12(e) shows an example composition of scatterplot andbar graph visualizations based on this design patter. In the figure,the scatterplot visualization is acting as a host and bar graph visu-alizations are nested inside its visual marks.
There is probably not a clear winner among different design pat-terns while designing an information visualization tool. The correctchoice of design pattern to use for a particular implementation de-pends on different conditions, such as the available view space, userknowledge, and the complexity of the underlying dataset. Ideallyspeaking, designers should be able to combine any existing visual-izations to generate a composite visualization view.
8.2 Delimitations
While our above CVV design patterns are general in nature, theyare based solely on the spatial layout of component visualizations.However, it is possible to envision other ways to combine two ormore visualizations, for example using interaction or animation.One such example is the use of interactive hyperlinking [6, 43] (orwormholing) to navigate between different visualization views.
8.3 Discussion
There are several direct benefits to structuring the design space ofcomposite visualization views in this manner. Classifying existingtechniques into patterns not only helps in understanding these tech-niques, but also in evaluating their strengths and weaknesses.
However, the design patterns presented in this paper are all basedon evidence from the literature of how existing visualization toolsand techniques use composite views. Therefore, our frameworkis inherently limited to current designs, and more descriptive thangenerative in nature. Furthermore, this list of patterns is not neces-sarily exhaustive, and we certainly foresee additional design pat-terns for composite views to emerge with progress in informa-tion visualization. It is also not always straightforward to sepa-rate what is a composite visualization and what is an “atomic” (orcomponent) visualization, particularly when the compositions onthe visual structures—which is the case for overloaded and nestedviews—as opposed to merely on the views. Our approach in theabove text has been to treat as components any technique has beenpresented in the literature as a standalone technique.
9 CONCLUSION
We have proposed a novel framework for specifying, designing, andevaluating compositions of multiple visualizations in the same vi-sual space that we call composite visualization views. The benefitof the framework is not only to provide a way to unify a large col-lection of existing work where visual representations are combinedin various ways, but also to suggest new combinations of visualrepresentations that may significantly advance the state of the art.
REFERENCES
[1] C. Ahlberg and B. Shneiderman. Visual information seeking: Tightcoupling of dynamic query filters with starfield displays. In Proceed-
Technique Visualization A Visualization B Spatial Relation Data Relation
ComVis [24] (Figure 2) any any juxtapose noneImprovise [39] (Figure 3) any any juxtapose noneJigsaw [36] any any juxtapose noneSnap-Together [30] any any juxtapose nonesemantic substrates [34] (Figure 4) node-link node-link juxtapose item-itemVisLink [11] (Figure 5) radial graph node-link juxtapose item-itemNapoleon’s March on Moscow [37] time line view area visualization juxtapose item-itemMapgets [38] (Figure 6) map text superimpose item-itemGeoSpace [22] (Figure 7) map bar graph superimpose item-item3D GIS [8] map glyphs superimpose item-itemScatter Plots in Parallel Coordinates [45] (Figure 8) parallel coordinate scatterplot overload item-dimensionGraph links on treemaps [14] (Figure 9) treemap node-link overload item-itemSparkClouds [21] tag cloud line graph overload item-itemZAME [13] (Figure 10) matrix glyphs nested item-groupNodeTrix [17] (Figure 11) node-link matrix nested item-groupTimeMatrix [44] matrix glyphs nested item-groupGPUVis [25] Scatterplot glyphs nested item-group
Table 1: Classification of common composite visualization techniques using our design space.
(a) Juxtaposed views. (b) Integrated views. (c) Superimposed views.1 2 3 4 5 6 7 8a b c d e f g h
(d) Overloaded views. (e) Nested views.
Figure 12: Example of composing a scatterplot and bar graph using different methods.
datasets in the same space and using different visualizations, butalso highlights the relational linking between the two datasets.
Nested views provide an efficient approach to link each of thedata values, visualized through the host visualization, to its relateddataset, visualized through client visualizations. This is achievedby nesting clients inside the visual marks in the host.
• Benefits: Very compact representation, easy correlation.• Drawbacks: Limited space for the client visualizations, clut-
ter is high, and visual design dependencies are high.• Applications: Again, situations that call for augmenting a
particular visual representation with additional mapping.
Figure 12(e) shows an example composition of scatterplot andbar graph visualizations based on this design patter. In the figure,the scatterplot visualization is acting as a host and bar graph visu-alizations are nested inside its visual marks.
There is probably not a clear winner among different design pat-terns while designing an information visualization tool. The correctchoice of design pattern to use for a particular implementation de-pends on different conditions, such as the available view space, userknowledge, and the complexity of the underlying dataset. Ideallyspeaking, designers should be able to combine any existing visual-izations to generate a composite visualization view.
8.2 Delimitations
While our above CVV design patterns are general in nature, theyare based solely on the spatial layout of component visualizations.However, it is possible to envision other ways to combine two ormore visualizations, for example using interaction or animation.One such example is the use of interactive hyperlinking [6, 43] (orwormholing) to navigate between different visualization views.
8.3 Discussion
There are several direct benefits to structuring the design space ofcomposite visualization views in this manner. Classifying existingtechniques into patterns not only helps in understanding these tech-niques, but also in evaluating their strengths and weaknesses.
However, the design patterns presented in this paper are all basedon evidence from the literature of how existing visualization toolsand techniques use composite views. Therefore, our frameworkis inherently limited to current designs, and more descriptive thangenerative in nature. Furthermore, this list of patterns is not neces-sarily exhaustive, and we certainly foresee additional design pat-terns for composite views to emerge with progress in informa-tion visualization. It is also not always straightforward to sepa-rate what is a composite visualization and what is an “atomic” (orcomponent) visualization, particularly when the compositions onthe visual structures—which is the case for overloaded and nestedviews—as opposed to merely on the views. Our approach in theabove text has been to treat as components any technique has beenpresented in the literature as a standalone technique.
9 CONCLUSION
We have proposed a novel framework for specifying, designing, andevaluating compositions of multiple visualizations in the same vi-sual space that we call composite visualization views. The benefitof the framework is not only to provide a way to unify a large col-lection of existing work where visual representations are combinedin various ways, but also to suggest new combinations of visualrepresentations that may significantly advance the state of the art.
REFERENCES
[1] C. Ahlberg and B. Shneiderman. Visual information seeking: Tightcoupling of dynamic query filters with starfield displays. In Proceed-
Juxtaposition
�5
collect such composite visualizations using literature searches andprior experience. We then let existing work inform our model byorganizing this prior art into rough categories that emerge from thecharacteristics of the techniques. In later sections, we discuss eachcategory in more detail. Finally, we construct a design space thatcaptures all aspects of these composite visualization techniques.
2.2 Visual Composition
The method for visual composition is an emerging theme when sur-veying composite visualizations in the literature. In other words,the different ways of composing two visualizations A and B in thesame visual space seems to be a useful organizing principle in thisdomain. Based on the literature, we derive the four visual compo-sitions (Figure 1) that give rise to four rough categories—we callthem CVV design patterns—for composing visualizations:
• Juxtaposition ! Juxtaposed Views: Placing visualizationsside-by-side in one view (Coordinated Multiple Views [32]);
• Superimposition ! Superimposed Views: Overlaying twovisualizations in a single view;
• Overloading ! Overloaded Views: Utilizing the space ofone visualization for another; and
• Nesting ! Nested Views: Nesting the contents of one visu-alization inside another visualization.
In addition, another emergent CVV design pattern is to juxtaposevisual structures, but to add graphical objects such as arrows, dottedlines, or glyphs to visually link one view with another. We thereforethink this method deserves a design pattern of its own:
• Integration ! Integrated Views: Placing visualizations inthe same view with visual links.
2.3 Design Patterns
Identifying and characterizing composite visualization views(CVVs) as a unified design approach not only allows us to explorethis space in a structured fashion, but also provides a method forcomparing the effectiveness of different designs. The reason weuse the term design pattern [15] here is that these are high-level ap-proaches where the actual composition generally differs on a case-by-case basis. This is consistent with the notion of a design patternas a general and reusable solution to a common problem.
We should also note that these design patterns are very differ-ent from the software design patterns for visualization proposed byHeer and Agrawala [16]. The latter deal with software engineeringdesign aspects, whereas our CVV patterns are defined on a visualdesign level. While the pattern movement is popular in softwareengineering, the reader should note that design patterns first wereproposed by Alexander et al. [2] for urban planning, and so our useof the concept is in fact closer to its original spirit.
Below we describe the five rough categories of composite visu-alization that we identified in the literature. In each section, wefirst describe each pattern and then give a couple of in-depth ex-amples of representative composite visualization techniques. Theseexamples are not intended to be exhaustive, but to be illustrative ofpractical implementations of each pattern.
2.4 Existing Formalisms
Using multiple views for visualization is not a new concept, andearly examples date back to the beginnings of the field [27]. Bal-donado et al. [4] gave general guidelines on the use of multi-ple views in information visualization, and North and Shneider-man [30, 28, 29] discussed relational models for achieving this.
Figure 2: ComVis [24] (Juxtaposed Views). Meteorology data.
Figure 3: Improvise [39] (Juxtaposed Views). Juxtaposed views are
used to explore the simulated ion trajectory in a cubic ion trap.
These discussions were later formalized into the concept of coor-dinated multiple views (CMV) [31, 32], where multiple views ofdifferent visualizations are combined in visual space and are im-plicitly linked together, often using brushing [5].
In their work on multiple and explicitly linked visualizations,Collins et al. [11] discuss the formalization of multi-relation visu-alizations, in the process deriving three different techniques for thispractice. Their formalism is related to our work but of a preliminarynature, lacks the discussion of some of the design patterns discussedhere, and also does not identify CVVs as a unified approach.
3 JUXTAPOSITION ! JUXTAPOSED VIEWS
Juxtaposed views (Figures 2 and 3) are the most prominent—andprobably the most flexible and easy to implement—design patternfor composing visualizations in a single view [4, 28, 31, 33]. Thedesign pattern is based on juxtaposing multiple visualizations sideby side. Any linking between visualizations is implicit, i.e., it is nota part of the visual representation. Examples include brushing [5],synchronized scrolling [27], and synchronized drill-down [23].
The effectiveness of juxtaposed views has been an important re-search topic. North and Shneiderman presented a taxonomy [29] ofsuch visualization. They showed that a well-designed juxtaposedview increases user performance while exploring relations amongmultiple data dimensions. However, designing effective juxtaposedviews can be a challenging task and requires efficient relationallinking and spatial layout. Weaver’s cross-filtered views [41] ad-dresses this by abstracting the relations between the views to makedefinining, implementing, and reusing them easier.
There currently exists a large number of visualization tools basedon juxtaposed views in the literature; e.g. [3, 7, 36]. Below wereview two such tools that are representative of these.
[ComVis, K. Matkovic et al., 2008]D. Koop, CIS 468, Fall 2018
Integration
�6
[Napoleon's March to Moscow, C. J. Minard, 1869]D. Koop, CIS 468, Fall 2018
"best statistical graphic ever"
(later known as a Sankey Diagram)
Multiple Views
�7
Facet
Partition into Side-by-Side Views
Superimpose Layers
Juxtapose and Coordinate Multiple Side-by-Side Views
Share Data: All/Subset/None
Share Navigation
All Subset
Same
Multiform
Multiform, Overview/
Detail
None
Redundant
No Linkage
Small Multiples
Overview/Detail
Linked Highlighting
[Munzner (ill. Maguire), 2014]D. Koop, CIS 468, Fall 2018
Multiple Views
�8
Facet
Partition into Side-by-Side Views
Superimpose Layers
Juxtapose and Coordinate Multiple Side-by-Side Views
Share Data: All/Subset/None
Share Navigation
All Subset
Same
Multiform
Multiform, Overview/
Detail
None
Redundant
No Linkage
Small Multiples
Overview/Detail
Linked Highlighting
[Munzner (ill. Maguire), 2014]D. Koop, CIS 468, Fall 2018
Multiform
�9
[Improvise, Weaver, 2004]D. Koop, CIS 468, Fall 2018
Overview-Detail View
�10
[Wikipedia]D. Koop, CIS 468, Fall 2018
Small Multiples• Same encoding, but different data in each view (e.g. SPLOM)
�11
[M. Bostock]D. Koop, CIS 468, Fall 2018
5 6 72.0 2.5 3.0 3.5 4.01 2 3 4 5 60.5 1.0 1.5 2.0 2.5
5
6
7
2.0
2.5
3.0
3.5
4.0
1
2
3
4
5
6
0.5
1.0
1.5
2.0
2.5
sepal length
sepal width
petal length
petal width
Multiple Views
�12
Facet
Partition into Side-by-Side Views
Superimpose Layers
Juxtapose and Coordinate Multiple Side-by-Side Views
Share Data: All/Subset/None
Share Navigation
All Subset
Same
Multiform
Multiform, Overview/
Detail
None
Redundant
No Linkage
Small Multiples
Overview/Detail
Linked Highlighting
[Munzner (ill. Maguire), 2014]D. Koop, CIS 468, Fall 2018
Partitioning: Matrix Alignment
�13
[Becker et al., 1996]D. Koop, CIS 468, Fall 2018
VISUAL DESIGN AND CONTROL OF TRELLIS DISPLAY 125
I page. In Figure 2 there are 6 panels, I column, 6 rows, and 1 page. Later, we willshow a Trellis display with more than one page. We refer to the rectangular array as thetrellis because it is reminiscent of a garden trelliswork .•
Each panel of a trellis display shows a subset of the values of panel variables;these values are formed by conditioning on the valqes of conditioning variables. In Fig-ure I the panel variables are variety and yield, and the conditioning variables are site andyear. On each panel, values of yield and variety are displayed for one combination of year
20 30 40 50 60
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchunaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475
Svansota
Waseca Waseca1932 1931
Crookston Crookston1932 1931
Morris Morris.i 1932 19:31
University Farm University Farm1932 1931
Duluth Duluth1932 1931
Grand Raoids Grand Raoids1932 1931
20 30 40 50 60
Barley Yield (bushels/acre)Figure 1. A Dotplot of the Barley Data Showing Yield Against Variety Given Year and Site.
Dow
nloa
ded
by [5
0.14
8.12
.36]
at 2
1:36
08
Janu
ary
2013
126 R. A. BECKER, W. S. CLEVELAND, AND M.-J. SHYU
1932 o 1931
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475Manchuria
No. 462Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatland
VelvetNo. 475
ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
TrebiWisconsin No. 38
No. 457GlabronPeatlandVelvet
No. 475ManchuriaNo. 462
Svansota
I I I IWaseca
0I:)0
000
00
00
Crookston00
00000
00
0Morris
00000
00
00
0University Farm
00
00
00
00
00Duluth
000
00
00
00
0Grand Rapids
00
00
00
00
00
20 30 40 50 60
Barley Yield (bushels/acre)Figure 2. A Dotplot of the Barley Data Showing Yield Against Site and Year Given Variety.
Dow
nloa
ded
by [5
0.14
8.12
.36]
at 2
1:36
08
Janu
ary
2013
Partitioning: Recursive Subdivision
�14
HillingdonEaling
Semi
Ter
Cam
den
Flat
Sem
i
Ter
Sem
i
Fla
t
Det
Southwark
Ter
Ter
Ter
Flat
Flat
Fla
t
Flat
Wandsw
ort
h
Sem
i
Det
Semi
Ter
Ter
Kin
gsto
n
Ter
Det
Det
Fla
t
Semi
Ter
Ter
Ter
Ter
Ter
Semi
Hounslo
w
Lewisham
Ter
Semi
Sem
i
Det
Redbridge
Barnet
Ter
Waltham Forest
Fla
t
Semi
Semi
Ter
Det
Richmond
Sem
i
Det
Croydon
Fla
t
Havering
GreenwichFlat
Ter
Det
Det
Flat
Flat
Mert
on
Ter
Flat
Fla
t
Det
Bro
mle
y
Flat
City of London
Ter
Fla
t
Ter
Haringey
Ter
Flat
Flat
Enfield
Flat
Fla
t
Brent
Fla
t
Ter
Fla
t
Ter
Ter
Sutton
Flat
Ter
Ter
Semi
Tower Hamlets
Ter
Islin
gto
n
FlatDet
Lambeth
Fla
t
Flat
Ter
Fla
t
Sem
i
Newham
Semi
Semi
Sem
i
Det
Ter
Flat
Westminster
Det
Semi
Sem
i
Semi
Ter
Sem
i
Semi
Sem
i
Semi
Semi
Fla
t
Ter
Kensington
Fla
t
Barking
Flat
Hack
ney
Flat Ter
Flat
Ham
mers
mith
Semi
Sem
i
Det
Harrow
Bexley
Semi
Flat
Southwark
Semi
Flat
Fla
t
Ter
Sem
i
Det
Ter
Lewisham
Fla
t
Sem
i
Semi
Flat
Det
Ter
Croydon
Richmond
Enfield
Flat
Ter
Ter
Greenwich
Barn
et
Ter
Ter
Flat
City o
f London
TerSemi
Sem
i
Flat
Fla
t
Hillingdon
Fla
t
Merton
Lambeth
Det
Semi
Det
Det
Flat
Ter
Sem
i
Sem
i
Flat
Haringey
Ter
Det
Fla
t
Ter
Ter
Ter
Semi
Harrow
Sem
i
Hounslow
Semi
Ter
Semi
Flat
Ter
Ter
Kingston
Waltham Forest
Sutton
Ter
Ter
Det
Kensington
Flat
Det
Bexley
Fla
t
Sem
i
Bromley
Ter
Westminster
Sem
i
Tower Hamlets
Ter
Ter
Semi
DetDet
Det
Fla
t
Ter
Fla
t
Flat
Flat
Ter
Fla
tTer
Ter
Flat Semi
Hackney
Sem
i
Ter
Det
Flat
Camden
Det
Semi
Ter
Newham
Ter
Barking
Semi
Det
Fla
t
Sem
i
Flat
FlatFlat
Sem
i
Det
Redbridge
Flat
Flat Sem
i
Ter
Semi
Fla
t
Ter
Semi
Flat
Det
Havering
Fla
t
Semi
Islington
Hammersmith
Ter
Wandsw
ort
h
Semi
Semi
Sem
i
Brent
Ealin
g
Fla
t
Ter
A B
Flat
Ter
Semi
Ter
Flat
Flat
Ter
FlatKensington
Det
Flat
Semi
Flat
Ter
Newham
Det
Ter
Det
Semi Det
TerFlat
Ter
Ter
Ter
Ter
Flat
Southwark
EalingTer
Semi
TerFlat Flat
Hounslow
TerIslington
DetKingston
Semi
Flat
Ter
Ter
Ter
DetSemi
Ter
Det
Lambeth
Ter
Semi
Haringey
Hillingdon
Semi
Flat
Ter
Det
Ter
Ter
Det
Waltham Forest
Ter
Ter
Merton
TerBarking
Flat
Det
Det
Semi
Det
Det
Det
Ter
Ter
SemiSemi
SemiSemi
Semi
Flat
Richmond
Semi
Westminster
Semi
Barnet
Ter
Det
Flat
City of London
Ter
Semi
Semi
Enfield
Wandsworth
Flat
Semi
Det
Flat
Flat
Redbridge
Flat
Ter
Flat
Det
Havering
Flat
Det
Semi
FlatTower Hamlets
Det
Semi
Semi
Det
Det
Det Det
Semi
Semi
Flat
Det
Semi
Lewisham
Camden
FlatFlat
Flat
Semi
Det
Semi
Flat
Croydon
Flat
Det
Brent
Flat
Semi
Det
Flat
Ter
Greenwich
Ter
Semi
Flat
Harrow
Det
SuttonSemi
Flat
Hackney
Semi
Bromley
Det
Semi
Ter
Hammersmith
TerBexley
Det
Semi
Det
Flat
Det
Det
Semi
Semi
Semi
Ter
Sutton
Ter
Havering
BexleyRichmond
Det
Ter
Brent
Ter
Semi
Wandsworth
Det
Flat
Flat
Barnet
Semi
Ter
Det
Hackney
Det
Semi
Flat
Flat
Semi
Kingston
Flat
Det
Det
Ter
Flat
Ealing Hammersmith
Ter
TerFlat
Flat
Det
Det
Ter
Det
Ter
Ter
Det
Ter
Det
Semi
Ter
Ter
Semi
Ter
Semi
Ter
Semi
Flat
Flat
Ter
Det Det
CamdenTer
Semi
FlatTer
Ter
Ter Flat
Det
Ter
Flat
Det
Flat
Flat
Flat
Flat
Det
Semi
Flat
DetSemi
Flat
Ter
Semi
Flat
Lambeth
Ter
Semi
Ter
Ter
Det
Flat
DetEnfield
Det
Westminster
Flat
DetHaringey
Semi
Det
Croydon
Flat
Semi
Flat
Lewisham
Det
Islington
Ter
Det
Semi
Southwark
Det
DetDet
FlatFlat
Flat
Redbridge
SemiSemi Semi
Flat
Semi
Ter
Det
Det
Hounslow
Ter
Det SemiSemi
Ter
Bromley
Flat
Ter
Barking
Det
Flat
Semi
Det
City of London
Semi
Merton
Flat
Semi
Semi
Ter
Semi
Greenwich
Waltham Forest
Kensington
Hillingdon
Ter
Tower Hamlets
Newham
Flat
Det
Semi
Harrow
Semi
Semi
Flat
C D
Fig. 2. A: Sized-based ordering, coloured by average price: sHier(/,$br,$ty,$yr,$mn); sLayout(/,SQ); sSize(/,$sal);sColor(/,Ø,Ø,Ø,$prc). B: Reconfigure to a spatial and temporal layout: oLayout(/,1,SP); oLayout(/,2,OS); oLayout(/,3,VT);oLayout(/,4,HZ). C: Fix the size: oSize(/,1,FIX); oSize(/,2,FIX); oSize(/,3,FIX); oSize(/,4,FIX). D: Remove time, andcolour by deviation from expected sales: oCut(/,4); oCut(/,3); oColor(/,2,$xsl).
the hierarchy can produce layouts similar to mosaic plots (and ma-trix diagrams if sizes are fixed). They are particularly suitable wherevariables have hierarchical dependencies, such as our calendar views(sHier($yr,$mn)).
6.3 Layouts for time-based data and questionsTemporal data can be considered as ordinal. In Fig. 1A, years arenot arranged temporally; as such, temporal trends are difficult to de-tect. Rearranging the years into a time-based order using an orderedspace-filling layout [36] (Fig. 1B) makes the increase in annual houseprice easier to detect. In Fig. 1C, we have added month to the hi-erarchy producing calendar views coloured by the number of sales.Seasonal variations in the numbers of sales are apparent for flats andterraced housing, however colour rescaling (using oColorMap) orusing colour schemes that are local to individual parts of the hierarchyare required to detect these patterns where property types have lowsales. Alternatively, colour can be used to show values as a proportionor deviation from a baseline. Appropriate baselines include those thatreflect the values expected from hypotheses that we might then acceptor reject on the basis of the display. For example, in Fig. 4A (calendarviews), our null hypothesis is that the number of sales does not varymonthly (expected or baseline values are a twelfth of the sales for eachyear). The geographically-consistent seasonal trends that are apparentmight cause us to reject our null hypothesis. Identifying the elementswith statistically-significant levels of variation might help us make that
choice. Fig. 4B shows the deviation of price from the yearly average(accounting for inflation). Whilst prices rises steadily every year, thisis not the case for 2008 where prices have dropped markedly in thefinal quarter, a trend not observed in Westminster.
Nesting the two temporal resolutions of year and month to pro-duce calendar views is appropriate where we are expecting yearly andmonthly patterns. However, this may obscure other temporal patterns.In Fig. 3B, we use an ordered squarified layout of all 108 months inthe period ordered from the left top to bottom right (compare with thecalendar views in Fig. 3A). Although both graphics show exactly thesame data, the use of $my and the associated OS layout in Fig. 3Bmake the upward trend in prices and subsequent slump more apparentas it is a continuous trend over the entire period. The result is a moreappropriate layout for research questions that relate to ongoing ratherthan periodic change. The additional hierarchical level used in Fig.3A and alternative layouts are more appropriate for comparing annualpatterns which are overshadowed by the longer term trend in the caseof this attribute. Again, interactive colour rescaling or colouring onthe basis of relative values is required to detect relative rises and fallsin different boroughs.
6.4 Geographical layoutsSpatially-ordered layouts (SP) have rectangles that are arranged ac-cording their geographical locations. The effect of this layout can beseen by comparing the non-spatial layout in Fig. 2A with the spatial
[Slingsby et al., 2009]D. Koop, CIS 468, Fall 2018
Assignment 4• http://www.cis.umassd.edu/~dkoop/cis468-2018fa/
assignment4.html • Interaction, Network, and Multiple Views • Due before Thanksgiving, but remember Quiz 2!
�15D. Koop, CIS 468, Fall 2018
Superimposition
�16Figure 6: Mapgets [38] (Superimposed Views). Presentation stack,
with superimposed layers for rivers, borders, and labels, in Mapgets.
Figure 7: GeoSpace [22] (Superimposed Views). A crime data layer
superimposed on a geographical map of the Cambridge, MA area.
5 SUPERIMPOSITION ! SUPERIMPOSED VIEWS
Superimposed views overlay two or more visual spaces on top ofeach other (Figures 6 and 7). The resulting visualization becomesthe visual combination of the component visualizations, often usingtransparency to enable seeing all views. Superimposed views aregenerally used to highlight spatial relations in the component visu-alizations. In other words, the spatial linking present in these viewsis one-to-one, i.e., all the overlay visualizations share the same un-derlying visual space. Line graph visualizations with several dataseries, where more than one graph is superimposed in a single chart(e.g., [19]), is a very commonly used example of this design pattern.
The spatial linking in the superimposed views allows for easycomparison across different datasets because the user does not haveto split their attention between different parts of the visual space.Furthermore, the fact that visualizations are stacked means that theycan each use the full available space in the view. However, becausethe composition simply adds the component visualizations together,the visual clutter may become significant, and it is also likely tocause conflicts arising from one visualization occluding another.
5.1 Mapgets
Mapgets [38] is a geographic visualization system that allows usersto interactively perform map editing and querying of geographicaldatasets. The maps generated using Mapgets are built on an under-lying presentation stack that superimposes multiple dataset layerson top of each other. The users can dynamically select the dataset
to use for each layer and the total number of layers to compose.Different layers in the presentation stack allow users to indepen-dently interact with each of the associated visualization and controlthe layer attributes. The technique also allows the users to reorderlayers in the presentation stack to achieve the desirable map result.
Figure 6 shows an example of a European map generated inMapgets. The presentation stack associated with this map consistsof three layers: the bottom layer visualizes rivers, the center layeris used to depict the country borders, and the topmost layer is usedto display the country labels.
5.2 GeoSpace
GeoSpace [22] allows users to interactively explore complex visualspaces using superimposed views. It permits progressively overlay-ing different datasets, based on the user queries, in a single view.Beyond allowing users to explore datasets through dynamic queries,GeoSpace also supports pan and zoom operations for navigation.
Figure 7 shows GeoSpace system being used for exploring crimearound the Cambridge, MA area. The figure shows a 2D view ofthe visualization, where red dots that are spatially coupled to theunderlying layer show the reported crime cases in the region.
Figure 8: SPPC [45] (Overloaded Views). This tool overloads points
into the region bounded by two axes in the parallel coordinate plot.
Figure 9: Links on treemaps [14] (Overloaded Views). The tool
identifies a tree structure in a graph and visualizes it using a treemap.
[Mapgets, A. Voisard, 1995]D. Koop, CIS 468, Fall 2018
Superimposed Layers• Put different layers in the same spatial region, overlay information • Usually each layer spans the entire view • Must be identifiable: visually distinguishable • Cartography has to deal with this a lot • May be static or dynamic (user controls which layers are shown)
�17D. Koop, CIS 468, Fall 2018
Example: Superimposed Line Charts
�18
October November December 2012 February March April May June July August September20
30
40
50
60
70
80
Tem
pera
ture
(ºF)
New York
San Francisco
Austin
[M. Bostock]D. Koop, CIS 468, Fall 2018
Example: Hierarchical Edge Bundles
�19
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 12, NO. 5, SEPTEMBER/OCTOBER 2006
Fig. 13. A software system and its associated call graph (caller = green, callee = red). (a) and (b) show the system with bundling strength β = 0.85using a balloon layout (node labels disabled) and a radial layout, respectively. Bundling reduces visual clutter, making it easier to perceive theactual connections than when compared to the non-bundled versions (figures 2a and 11a). Bundled visualizations also show relations betweensparsely connected systems more clearly (encircled regions); these are almost completely obscured in the non-bundled versions. The encircledregions highlight identical parts of the system for (a), (b), and figure 15.
Fig. 14. Using the bundling strength β to provide a trade-off between low-level and high-level views of the adjacency relations. The value of βincreases from left-to-right; low values mainly provide low-level, node-to-node connectivity information, whereas high values provide high-levelinformation as well by implicit visualization of adjacency edges between parent nodes that are the result of explicit adjacency edges between theirrespective child nodes.
changing the bundling strength β and by switching between differ-ent tree layouts. The participants from academia were our fellow re-searchers, PhD students and MSc students from the Computer Sciencedepartment of the Technische Universiteit Eindhoven. They all had ex-perience with either software development, software visualization, orinformation visualization in general. Participants from industry wererepresentatives of the Software Improvement Group (SIG) in Amster-dam, which delivers insight in the structure and technical quality ofsoftware portfolios, and representatives of FEI Company Eindhoven,which produces software to operate with FEI’s range of electron mi-croscopes.
The majority of the participants regarded the technique as usefulfor quickly gaining insight in the adjacency relations present in hier-archically organized systems. In general, the visualizations were alsoregarded as being aesthetically pleasing. SIG and FEI Company Eind-hoven are currently supporting further development by providing uswith additional data sets and feedback regarding the resulting visual-izations.
More specifically, most of the participants particularly valued thefact that relations between items at low levels of the hierarchy wereautomatically lifted to implicit relations between items at higher lev-els by means of bundles. This quickly gave them an impression of thehigh-level connectivity information while still being able to inspectthe low-level relations that were responsible for the bundles by inter-actively manipulating the bundling strength.
This is illustrated in figure 14, which shows visualizations usingdifferent values for the bundling strength β . Low values result in vi-sualizations that mainly provide low-level, node-to-node connectivityinformation. High values result in visualizations that provide high-level information as well by implicit visualization of adjacency edgesbetween parent nodes that are the result of explicit adjacency edgesbetween their respective child nodes.
Another aspect that was commented on was how the bundles gavean impression of the hierarchical organization of the data as well,thereby strengthening the visualization of the hierarchy. More specif-ically, a thick bundle shows the presence of two elements at a fairly
746
[D. Holten, 2006]D. Koop, CIS 468, Fall 2018
D3 Multiple Views and Interaction• https://codepen.io/dakoop/pen/oQxxmx • Process mouseover and mouseout events
- Get selected element - Provide feedback (e.g. highlighting)
• Find matching items in other view(s) - Can use filter for this - Highlight them - Make sure that if they overlap, the highlighted item is on top
�20D. Koop, CIS 468, Fall 2018