Exam: kartičky | information-visualization

InfoVis × SciVis × DataVis

InfoVis – use of computer-supported, interactive visual representations of data to amplify cognition
SciVis – datasets have a given spatialisation
DataVis – web-based, communication
challenges: diversity, scale

visual information-seeking mantra

visual perception is a two stage process

Gestalt psychology

one stimulus, two perceptions
there is a difference between stimulus and perception
emergence – we need to see the whole picture, not just parts
reification – perception contains more spatial information than the stimulus
multistability – ambiguous stimuli can generate different perceptions but they cannot coexist simultaneously
invariance – objects are recognized independently of various variations (transformations, lightning, …)
laws of grouping – law level perceptions are grouped into higher-level objects
- good Gestalt

information visualization pipeline

source data → data tables
- data transformations
data tables → visual abstraction
- visual mappings
- transition from data form to visual form
- we use data attributes to create (visual) marks
visual abstraction → views
- view transformations
- result … actual pixels

taxonomies of data types

“what comparisons can I make?”
“how can I aggregate the data?”
nominal … only equality (=)
- aggregation … mode (we show the most frequent one) or top-k
  - if we have a taxonomy (hierarchy), we can group the data
ordered … ordering and equality (<, =)
- aggregation … median, quantiles, histogram (count per bucket)
quantitative … “how much smaller is it?”
- → intervals … $v-v'$
- → ratios … $v/v'$ $v / v^{'}$
  - only if there is a meaningful zero on the scale
- aggregation … mean, std dev, skew

properties of visual channels

association $(\equiv)$ $(\equiv)$
- does the visual channel play well with the other visual channels?
- size does not provide association – the other visual variables are more difficult to see for smaller sizes
selection $(\neq)$ $(\neq =)$
- can you focus on a specific subset in this visual channel?
- color provides selection
order $(O)$ $(O)$
- items can be ordered according to this variable, without relying on a lookup to a legend
quantity $(Q)$ $(Q)$
- the difference between two items can be quantified

channels (according to Bertin)

position $(\equiv,\neq,O,Q)$
size $(\neq,O,Q)$ $(\neq =, O, Q)$
- beware: for quantity judgement, our perception is biased
  - Stevens' Law: $\frac{p(x_1)}{p(x_0)}=\left(\frac{x_1}{x_0}\right)^\beta$
  - $p$ … perception
  - $\beta$ is different for length, area, and volume
  - in 2D, we underestimate large sizes (in 3D it's even worse)
- note on depth
  - depth is perceived mainly because it impacts size
  - using the third dimension introduces ambiguity: is it small or is it far away?
  - but we can keep all the objects the same size and only apply motion parallax & skew instead of perspective
value $(\neq,O)$ $(\neq =, O)$
- lightness of color
- light marks are harder to see
texture $(\equiv,\neq,O)$ $(\equiv, \neq =, O)$
- usually black and white (dark and light color) – we can mix it with color channel
- probably underused
color $(\equiv,\neq)$ $(\equiv, \neq =)$
- we don't perceive the axis of wavelengths
- colors are usually used as labels
orientation $(\equiv,\neq)$ $(\equiv, \neq =)$
- only for some shapes (not circles)
shape $(\equiv)$ $(\equiv)$
- does not provide selection → it's better to use color for grouping

Mackinlay: suitability of variables, possible combinations

Card & Mackinlay table

interesting visualizations

multi-dimensional data
- scatter plot matrix
- parallel coordinates
time series
- horizon graph – example of composite visual mapping (one attribute → multiple graphic variables)