2 December 2005 Information Visualisation Data Presentation Prof. Beat Signer Department of Computer Science Vrije Universiteit Brussel beatsigner.com

Beat Signer - Department of Computer Science - [email protected] 2 March 14, 2024 Information Visualisation Process Data Representation Data Data Presentation Interaction mapping perception and visual thinking

Beat Signer - Department of Computer Science - [email protected] 3 March 14, 2024 Marks and Channels ▪ Marks are basic graphical elements (geometric primi- tives) to represent items or links ▪ Channels control the appearance of marks ▪ Vis design space described by orthogonal combination of marks and channels ▪ Complex visual encodings can be decomposed and analysed in terms of their marks and channels ▪ building blocks for analysing visual encodings

Beat Signer - Department of Computer Science - [email protected] 4 March 14, 2024 Marks ▪ Basic geometric/graphical element in an image ▪ classified according to the number of spatial dimensions

Beat Signer - Department of Computer Science - [email protected] 5 March 14, 2024 Marks … ▪ Zero-, one- or two-dimensional marks ▪ three-dimensional marks (volumes) are not used frequently due to limited perception

Beat Signer - Department of Computer Science - [email protected] 6 March 14, 2024 Mark Types ▪ Item marks ▪ Link marks ▪ connection marks - pairwise relationship between two items via a line ▪ containment marks (enclosure or nesting) - hierarchical relationships using areas

Beat Signer - Department of Computer Science - [email protected] 7 March 14, 2024 Channels

Beat Signer - Department of Computer Science - [email protected] 8 March 14, 2024 Channels … ▪ Control appearance of mark independently of the dimensionality of the geometric primitive ▪ Many visual channels ▪ spatial position ▪ shape ▪ colour (hue, saturation and luminance) ▪ motion (e.g.flicker, direction and velocity) ▪ size (i.e.length, area and volume) ▪ tilt (angle) ▪ Size and shape channels cannot be used on all types of marks ▪ e.g. area marks typically not size or shape coded

Beat Signer - Department of Computer Science - [email protected] 9 March 14, 2024 Channel Types ▪ Identity channels ▪ information about what something is ▪ e.g.shape, hue colour channel, motion pattern ▪ Magnitude channels ▪ how much of something is there ▪ e.g.size (length, area or volume), luminance or saturation colour channels, angle, …

Beat Signer - Department of Computer Science - [email protected] 10 March 14, 2024 Using Marks and Channels ▪ Progression of chart types ▪ (a) one quantitative and one categorical attribute ▪ (b) scatterplot with two quantitative attributes ▪ (c) two quantitative and one categorical attribute via hue ▪ (d) three quantitative (one via size) and one categorical attribute ▪ In examples each attribute encoded via single channel ▪ multiple channels might also be used redundantly

Beat Signer - Department of Computer Science - [email protected] 11 March 14, 2024 Using Marks and Channels … ▪ Use of marks and channels guided by the principles of expressiveness and effectiveness ▪ after identifying most important attributes ensure that they are encoded with the highest ranked channel ▪ Expressiveness principle ▪ visual encoding should express all of, and only, the information in the dataset attributes - ordered data (ordered attributes) should be shown in a way that our perceptual system senses as ordered → use magnitude channels - unordered data (categorical attributes) should not be shown in a way that perceptually implies an ordering that does not exist → use identity channels

Beat Signer - Department of Computer Science - [email protected] 12 March 14, 2024 Using Marks and Channels … ▪ Effectiveness principle ▪ importance of attribute should match the salience of the channel ▪ most important attributes (depends on the task) encoded with most effective channels ▪ Attributes encoded with position will dominate the user's mental model ▪ choice of which attributes to encode with position is the most central choice in visual encoding

Beat Signer - Department of Computer Science - [email protected] 13 March 14, 2024 Channel Effectiveness [Visualization Analysis & Design, Tamara Munzner, 2014]

Beat Signer - Department of Computer Science - [email protected] 14 March 14, 2024 Channel Effectiveness … ▪ Obvious way to quantify effectiveness via accuracy ▪ how close is human perceptual judgement to some objective measurement of the stimulus? ▪ Different visual channels are perceived with different levels of accuracy ▪ characterised by Steven's Psychophysical Power Law

Beat Signer - Department of Computer Science - [email protected] 15 March 14, 2024 Steven's Psychophysical Power Law ▪ Responses to sensory experience of magnitude are characterisable by power laws ▪ 𝑆 = perceived sensation ▪ I = physical intensity ▪ exponent N depends on sensory modality ▪ most stimuli are magnified (superlinear) or compressed (sublinear) [Visualization Analysis & Design, Tamara Munzner, 2014]

Beat Signer - Department of Computer Science - [email protected] 16 March 14, 2024 Error Rates Across Channels Results by Cleveland and McGill, 1984

Beat Signer - Department of Computer Science - [email protected] 17 March 14, 2024 Channel Effectiveness … ▪ Channel effectiveness mainly based on accuracy but also takes into account ▪ discriminability ▪ separability ▪ popout ▪ grouping

Beat Signer - Department of Computer Science - [email protected] 18 March 14, 2024 Discriminability ▪ Quantify the number of distinguishable steps (bins) that are available within a visual channel ▪ some channels (e.g.line width) have a very limited number of bins ▪ small number of bins is not a problem if the number of values to be encoded is also small ▪ number of different values that need to be shown for an attribute must not be greater than the available bins for the visual channel - otherwise aggregate or use different visual channel

Beat Signer - Department of Computer Science - [email protected] 19 March 14, 2024 Effective Line Width Use ▪ Limited number of discriminable bins ▪ line width works well for 3 or 4 different values

Beat Signer - Department of Computer Science - [email protected] 20 March 14, 2024 Separability ▪ Channels are not always completely independent from each other (interchannel interference) ▪ ranging from fully separable channels to the inseparably combined integral channels (major interference) ▪ Visual encoding straightforward with separable channels ▪ encoding of different information in integral channels will fail

Beat Signer - Department of Computer Science - [email protected] 21 March 14, 2024 Popout ▪ Many channels provide visual popout (preattentive processing) where a distinct item stands out from many others immediately ▪ time to spot the different object does not depend on the number of distractor objects (a) vs.(b) ▪ massively parallel processing of low-level features ▪ popout effect slower for shapes ((c) and (d)) than for colour hue channel ((a) and (b)) ▪ channels with individual popout cannot simply be combined ((e) and (f)) - need serial search to find the red circle in (f) ▪ Most pairs of channels do not support popout ▪ use popout for a single channel at a time

Beat Signer - Department of Computer Science - [email protected] 22 March 14, 2024 Popout …

Beat Signer - Department of Computer Science - [email protected] 23 March 14, 2024 Popout Channels ▪ Popout cannot only occur for colour hue and shape channels ▪ tilt (a) ▪ size (b) ▪ shape (c) ▪ proximity (d) ▪ shadow direction (e)

Beat Signer - Department of Computer Science - [email protected] 24 March 14, 2024 Popup Channels …

Beat Signer - Department of Computer Science - [email protected] 25 March 14, 2024 Grouping ▪ Containment (links) is the strongest cue for grouping followed by connection coming in second ▪ Items sharing the same level of a categorical attribute can also be perceived as a group ▪ Proximity is the third-strongest grouping approach ▪ Similarity (hue, motion and shape) ▪ shape and motion channel to be used with care

Beat Signer - Department of Computer Science - [email protected] 26 March 14, 2024 Relative versus Absolute Judgements ▪ Perceptual system fundamentally based on relative judgements and not absolute ones (Weber's Law) ▪ e.g.position along a scale can be perceived more accurately than pure length judgement without a scale

Beat Signer - Department of Computer Science - [email protected] 27 March 14, 2024 Relative Luminance Perception ▪ Perception of luminance is contextual based on the contrast with surrounding colours

Beat Signer - Department of Computer Science - [email protected] 28 March 14, 2024 Colour (Hue) Perception ▪ Our visual systems evolved to provide colour constancy ▪ same surface identifiable across illumination conditions ▪ visual system might work against simple colour encodings [Visualization Analysis & Design, Tamara Munzner, 2014]

Beat Signer - Department of Computer Science - [email protected] 29 March 14, 2024 Mapping Colour ▪ Luminance and saturation are magnitude channels while hue is an identity channel ▪ luminance can be used for two to four levels (bins) ▪ saturation can be used for up to three levels (bins) - strongly interacts with size channel ▪ saturation and hue are non- separable channels for small regions

Beat Signer - Department of Computer Science - [email protected] 30 March 14, 2024 Comparing HSL Lightness ▪ Computed HSL lightness L is the same for all six colours ▪ true luminance as measured by an instrument ▪ perceived luminance L* represents what we see - more sensitive to certain wavelengths (green and yellow) as shown earlier with the spectral sensitivity

Beat Signer - Department of Computer Science - [email protected] 31 March 14, 2024 No Implicit Order for Hue ▪ Sometimes learned hue orders (not at perception level) ▪ green-yellow-red from traffics lights ▪ rainbow colour ordering

Beat Signer - Department of Computer Science - [email protected] 32 March 14, 2024 Colourmaps ▪ A colourmap defines a mapping between colours and data values ▪ Colourmaps can be categorical or ordered (sequential or diverging) ▪ use magnitude channels of luminance and saturation for ordered data

Beat Signer - Department of Computer Science - [email protected] 33 March 14, 2024 Colourmap Categorisation (Taxonomy) [Visualization Analysis & Design, Tamara Munzner, 2014]

Beat Signer - Department of Computer Science - [email protected] 34 March 14, 2024 Categorical Colourmaps ▪ Categorical colourmaps (qualitative colourmaps) are normally segmented (not continous) ▪ effective for categorical data (next best channel after position) ▪ Good resource for creating colourmaps is ColorBrewer

Beat Signer - Department of Computer Science - [email protected] 35 March 14, 2024 Categorical Colourmaps … ▪ Can use six to twelve distinguishable hue steps (bins) for small separated regions ▪ includes background colour and default object colours ▪ use easy nameable colours: e.g.red, blue, green, yellow, orange, brown, pink, magenta, purple and cyan ▪ Use highly saturated colours for small regions ▪ Use low-saturation colours (pastels) for large regions

Beat Signer - Department of Computer Science - [email protected] 36 March 14, 2024 Ineffective Categorical Colourmap Use

Beat Signer - Department of Computer Science - [email protected] 37 March 14, 2024 Example of Using Additional Channels ▪ Dataset with 27 categorical levels from 7 categories

Beat Signer - Department of Computer Science - [email protected] 38 March 14, 2024 Example of Using Additional Channels …

Beat Signer - Department of Computer Science - [email protected] 39 March 14, 2024 Example of Using Additional Channels …

Beat Signer - Department of Computer Science - [email protected] 40 March 14, 2024 Example of Using Additional Channels …

Beat Signer - Department of Computer Science - [email protected] 41 March 14, 2024 Ordered Colourmaps ▪ Sequential colourmap ranges from a minimum value to a maximum value ▪ use luminance (with or without hue) or saturation channel ▪ Diverging colourmap ▪ use two hues at the endpoints and a neutral colour (e.g.white or grey) as a midpoint

Beat Signer - Department of Computer Science - [email protected] 42 March 14, 2024 Rainbow versus Two-Hue Colour Map ▪ How many hues to use in continous colourmaps? ▪ high-level structure versus local neighbourhoods (fine grained details) ▪ rainbow colourmap makes it easier to discuss specific (nameable) subranges

Beat Signer - Department of Computer Science - [email protected] 43 March 14, 2024 Rainbow Continous Colourmaps ▪ Problems of rainbow continous colourmaps ▪ hue is used to indicate order (despite being an identity channel) ▪ scale is not perceptually linear ▪ fine details cannot be perceived via the hue channel - luminance channel much better (luminance contrast required for edge detection in our eyes)

Beat Signer - Department of Computer Science - [email protected] 44 March 14, 2024 Rainbow Continous Colourmaps … ▪ The three problems of rainbow continous colourmaps can be addressed by using monotonically increasing luminance colourmaps ▪ multiple hues are ordered according to their luminance from lowest to highest ▪ Rainbow colourmap ▪ standard rainbow colour- map (a) vs.perceptually linear rainbows (b) with decreased dynamic range ▪ segmented rainbow for categorical data (c)

Beat Signer - Department of Computer Science - [email protected] 45 March 14, 2024 Bivariate Colourmaps ▪ Safest use of colour channel is to visually encode a single attribute (univariate) ▪ In the colourmap categorisation we have seen colourmaps encoding two separate attribute (bivariate) ▪ if one of the two attributes is binary then it is straightforward to create a comprehensible bivariate colourmap - choose base set of hues and vary the saturation ▪ if both attributes are categorical with multiple levels the results will be poor ▪ combinations of sequential and diverging attributes should be used carefully - appear frequently in vis solutions but some people have difficulties to interpret their meaning

Beat Signer - Department of Computer Science - [email protected] 46 March 14, 2024 Colourblind-Safe Colourmaps ▪ A safe strategy is to avoid using the hue channel only ▪ e.g.vary luminance or saturation in addition to hue in categorical colourmaps ▪ Avoid colourmaps emphasising red-green (e.g.diverging red-green colourmap) ▪ Use colour blindness simulators and tools such as Viz Palette

Beat Signer - Department of Computer Science - [email protected] 47 March 14, 2024 Size Channels ▪ Suitable for ordered data but interacts with most other channels ▪ length (1D) - judgment of length is very accurate ▪ area (2D) - judgement of area is less accurate ▪ volume (3D) - volume channel is quite inaccurate

Beat Signer - Department of Computer Science - [email protected] 48 March 14, 2024 Tilt/Angle Channel ▪ Encode magnitude information based on the orientation of a mark ▪ tilt: orientation against the global frame of the display ▪ angle: orientation of a line with respect to another line ▪ Accuracy of our perception of a tilt/angle is not uniform ▪ very accurate near exact horizontal, vertical or diagonal positions

Beat Signer - Department of Computer Science - [email protected] 49 March 14, 2024 Other Channels ▪ Shape channel ▪ commonly applied to point marks ▪ can also be applied to line marks (e.g. dotted or dashed lines) ▪ can distinguish between dozens up to hundreds bins - strong interaction between shape and size channel ▪ Motion channels ▪ direction of motion ▪ velocity of motion ▪ flicker and blinking frequency ▪ very separable from all other static channels ▪ strongly draws attention (strong popout) - hard to ignore and should be used carefully

Beat Signer - Department of Computer Science - [email protected] 50 March 14, 2024 Other Channels … ▪ Texture and stippling channel ▪ texture can be simplified by considering it as a combination of the following three perceptual dimensions - orientation, scale and contrast ▪ texture can be used to show categorical attributes as well as ordered attributes ▪ Stippling fills regions of drawings with short strokes - e.g. dashed or dotted lines - used for area marks in older printing (to simulate grey)

Beat Signer - Department of Computer Science - [email protected] 51 March 14, 2024 Exercise 5 ▪ Visualisation in Python

Beat Signer - Department of Computer Science - [email protected] 52 March 14, 2024 Further Reading ▪ This lecture is mainly based on the book Visualization Analysis & Design ▪ chapter 5 - Marks and Channels ▪ chapter 10 - Map Color and Other Channels

Beat Signer - Department of Computer Science - [email protected] 53 March 14, 2024 References ▪ Visualization Analysis & Design, Tamara Munzner, Taylor & Francis Inc, (Har/Psc edition), May, November 2014, ISBN-13: 978-1466508910 ▪ Semiology of Graphics: Diagrams, Networks, Maps, Jacques Bertin, ESRI PR (1st edition), January 2010, ISBN-13: 978-1466508910 ▪ Information Visualization: Perception for Design, Colin Ware, Morgan Kaufmann (3rd edition) May 2012, ISBN-13: 978-0123814647

Beat Signer - Department of Computer Science - [email protected] 54 March 14, 2024 References … ▪ ColorBrewer 2.0 ▪ https://colorbrewer2.org ▪ Viz Palette ▪ https://projects.susielu.com/viz-palette

2 December 2005 Next Lecture Data Processing and Visualisation Frameworks