Harvard CS171 - Visualization by Hanspeter Pfister

Lec.1 Introduction

What is visualization?

To convey information through graphical representations of data

Visualization goals:

• Record information
• Analyze data to support reasoning
  • Develop and assess hypotheses(visual exploration)
  • Find patterns and discover errors data
• Communicate information to others
  • Share and persuade(visual explanation)

Lec.2 Design Principles

Clutter and confusion are not attributes of information, they are failures of design - E.Tufte

Tufte’s Integrity Principles:

• Show data variation, not design variation
• Clear, detailed, and through labeling and approriate scales
• Size of the graphcis effect should be directly proportinal to the numerical quantities

Tufte’s Design Principles:

• Maximize data-ink ratio
• Avoid chart junk
• Increase data density
• Layering of information

Lec.3 Data Models & Visual Variables

Data Types:

• Nominal(Categorical)
  • Are = or $\neq$ to other values
• Ordinal
  • Obey a < relationship
• Quantitative
  • Can do arithmetic on them
  • Interval (no “true zero”): dates, location. Cannot compare directly. Only differences can be compared.
  • Ratio (zero fixed): Origin is meaningful, can measure ratios & proportions.

Data vs. Conceptual Model:

• Data Model: Low-level description of the data(1D floats, 3D vector of floats)
• Conceptual Model: Mental construction(temperature, space)

Bertin’s Visual Variables:

Objectives: Points, Lines, Areas

Channels: Position, Size, (Grey)Value, Texture, Color, Orientation, Shape

• Points: Strongest visual variable; Suitable for all data types
  • Sometimes not available
  • Cluttering
• Size & Length: Good visual variable; Easy to see whether one is bigger; Grouping works
  • Good for aligned bars
  • OK for changes in length
  • Bad for changes in area
• Shape: Great to recognize many classes
  • No grouping, ordering
• Value: Good for quantitative data when length & size are used; Supports grouping; Is preattentive if sufficiently different
  • Not very many shades recognizable
• Color: Good for qualitative data; Is preattentive if sufficiently different
  • Limited number of classes
  • Not good for quantitative data
  • Lots of pitfalls! Be careful!

Characteristics of Visual Variables:

• Selective: Is a mark distinct from other marks?
• Associative: Does it supporting grouping?
• Quantitative: Can we quantify the difference between two marks?
• Order: Can we see a change in order?
• Length: How many unique marks can we make?

Graphical Perception:

Lec.4 Interaction

The Shneiderman Mantra: Overview first, zoom and filter, and details on demand.

• Focus + Context
• Brushing & Linking
• Filtering
• Zooming
• Animation

Lec.5 Networks & Trees

TBD

Lec.6 Data Visualization Process & Graphs

• Target
  • Choose a specific domain
  • Define research question(s)
  • Find & clean the data
• Translate (What? 80%)
  • Formulate data analysis tasks
  • Exploratory data analysis
  • Transform & summarize data
• Design (How? 20%)
  • Design visual encodings
  • Design interactions
  • Sketch many ideas!
• Implement
  • Use code “sketches”
  • Define data structures
  • Find efficient algorithms
• Validate
  • Is the visualization effective?
  • Does it support the tasks?
  • Does it provide new insights?

Comparisions: Bar Chart, Waterfall Chart, Dot Plots

Bar Chart vs. Line Chart: Line implies trend. Do not use for categorical data.

Two line segments are maximally discriminable when their average absolute angle is $45^\circ$.

Trends Over Time: Line Chart, Streamgraphs

Proportions: Pie Chart, Donut CHart, Stacked Bar Chart, Small Multiples, Stacked Area Chart

Distribution: Histogram, Density Plots, Box & Whisker Plots

Correlations: Scatterplots, Trend Lines, Residual Graph, Quadrants, Path Plots

Visualization Taxonomy: What Makes a Visualization Memorable?

Lec.7 High-Dimensional Data

• How many dimensions?
  • ~50: tractable with “just” vis
  • ~1000 - need analytical methods
• How many records?
  • ~1000 - “just” vis is fine
  • >10,000 - need analytical methods
• Homogeneity?
  • Same data type?
  • Same scales?

Geometric Methods

• Parallel Coordinates: Shows primarily relationships between adjacent axis
  • Each axis represents dimension
  • Lines connecting axis represent records
  • Suitable for all tabular data typesa (Also heterogeneous data)
  • Limited scalability (~50 dimensions, ~1-5k records)
  • Interaction is crucial: Axis reordering, Brushing, Filtering
• Star Plot: Similar to parallel coordinates, radiate from a common origin
• Small Multiples: Use multiple views to show different partitions of a dataset
• Scatterplot Matrices(SPLOM): Matrix of size d*d, each cell plots a scatterplot of two dimensions
  • Limited scalability (~20 dimensions, ~500-1k records)
  • Brushing is important
  • Often combined with “Focus Scatterplot” as F+C technique
• Others: Connected Charts, Domino, etc.

Pixel Based Methods

• Each cell is a “pixel”, value encoded in color/value
• Meaning derived from ordering
• If no ordering inherent, clustering is used
• Good for homogeneous dataset
• Color is relative!(Also should consider the blindness)

Dimensionality Reduction

• Principal Component Analysis(PCA)
  • Linear mapping, by order of variance
• Multidimensional Scaling(MDS)
  • Nonlinear, better suited for some dataset
  • Popular for text analysis

Can we trust dimensionality reduction?: Interpretation and Trust: Designing Model-Driven Visualizations for Text Analysis

Glyph

Lec.8 Text and Document Visualization

Features of Text:

• Abstract
• General for mental concepts
• Different across population groups
• Linear perception
• Semi-structured
• Legibility!

Vis for Text Collections:

• TagClouds
• WordCloud
• The Word Tree
• Mapping Text with Phrase Nets
• DocuBurst(based on WordNet)

Vis for Large Document Collections:

• Bohemian bookshelf
• DocumentCards
• Semanticons

Others:

• Text in Vis Storytelling
• Time-Evolving Document Collections
• Storyline

Lec.9 Perception

Hallucination:

• Herman Grid Effect
• Cornsweet Illusion
• Others

Tips:

1. We are scanning visualizations in reading order and are attracted to titles, text, and labels. Put titles at the top left and put labels and textual explanations close to the visualization.
2. Our visual system sees differences, not absolute values, and is attracted to high-contrast edges. Maximize contrast between visual elements to make them stand out and avoid busy textures.
3. Draw the user’s eye to the most important part of the visualization. Provide a visual hierarchy of imnformation that will help make it clear to the user how they should interact with the information.
4. We can easily see objects that are different color or that are in motion. Use color and motion sparingly to make the important information pop out.
5. The brain is organized hierarchically, with higher brain regions processing more abstract features. User-recognizable objects in a visualization will be processed differently than low-level features.
6. Some channels are integral and cannot be separated without a lot of cognitive effort. Be careful whenever you use multiple channels to encode information.

Gestalt Principles

Proximity, Similarity, Connection, Enclosure, Grouping, Continuity, Closure, Symmetry, Figure/Ground, Common Fate.

Lec.10 Color

“… avoiding catastrophe becomes the first principle in bringing color to information: Above all, do no harm.” –E.R. Tufte

Additive Color Mixing: When colors combine by adding their spectra

• Monitors, Projectors, Slide Film, etc.

Subtractive Color Mixing: When colors combine by multiplying their spectra

• Light reflecting off a surface, optical filters, crayons, etc.

Visual Cones:

Color Perception:

Higher contrast sensitivity in luminance than in chrominance.

Bezold Sperading Effect

A Color-Caused Optical Illusion on a Statistical Graph

Color Blindness: - Red / Green deficiencies: No L cones or No M cones - Blue /Yellow deficiencies: No S cones

About 7-10% of the male population is red-green colorblind. Make sure that colors are not your only method of conveying important information.

Color Space:

• RGB Color Space:
  • Additive system
  • Colors that can be represented by computer monitors
  • Not perceptually uniform
• HSL Color Space
  • Hue - what people think of color
  • Saturation - distance from grey
  • Luminance - from dark to light
  • Not perceptually uniform - brightness of hues seems to vary
• LAB / HCL Color Spaces
  • Perceptually uniform!
  • L - Luminance
  • AB - approximate R/G and Y/B opponent channels
  • HCL: Hue-Chroma-Lightness: Cylindrical transformation of LAB
• Munsell Color Space
  • Perceptually uniform version of HSL
  • Surface colors, not emitted light
  • Used for paint swatches

Colors for Nominal Data:

• Ware suggests:
  • red, green, yellow, blue, black, white
  • pink, cyan, grey, orange, brown, purple

Nominal colors, sequential colors, diverging colors.

Lect.11 Cognition

Perception vs. Cognition:

• Perception: About the nature of the signals coming in; What you see
• Cognition: About how you understand and interpret what you see

Vision is “Constructed” top down from the input.

“What you see when you see a thing depends on what the thing is. What you see the thing as depends on what you know about what you are seeing.”

Image Gist

• From one glance to the next, the most of the visual information is a summary of what’s there, which lacks local detail.
  • The gist refers to the visual information perceived after/during a glance at an image
  • In a glance(~100ms), we remember the meaning of an image and its global layout
  • Some objects and details are forgotten

Tips:

• Image gist gives you layout
• Can give you general categorization info but lacks detail
• Atten is required to get specific details

Visual Attention

• Attentiion is Selective
  • Attentional blindness
  • Attention to one aspect or feature of the input suppresses processing of other features
• Where people look depends on knowledge
  • People learn something more easily if they can relate it to something they already know

Tips:

• To find meaning in what we see we must selectively pay attention to what is important
• Low-level vision is driven by object features rather than a conscious effort where to look(e.g., pre-attentive processing)
• Attention is driven by preexisting knowledge, expectations, and goals stored in long-term memory

Visual Working Memory

Georage Miller’s 7$\pm$2 rule; Cowen: 3-4 visual objects

The number of items that can be held in VSTM depends on their complexity.

Visual Long Term Memory

What we know: People can remember thousands of scenes.

What we don’t know: What people are remembering for each item?

• “Gist” Only? Sparse detail? Highly Detailed?

Hypothesises:

• People only remember the gist
  • Exemplars of the same category will interfere
• Efficient recognition; no interferenec
  • Representation of each image has a unique code that distinguishes itself from all the other images

What Makes a Visualization Memorable?

• It contains human distinguishable objects
• It is a distinct visualization type
• It is colorful
• It is visually dense
• It has a low data-to-ink ratio

Conclusion

• Gist
  • Spatial layout
  • Basic level category
  • Experience of rich percept
  • No local details
• Attention & eye movements
  • Where you look depends on: salience, task-demands, prior knowledge
  • Attending to specific areas fills in local details
• Working memory
  • Limited capacity
  • Amount of information depends on experience and complexity
  • Eye movements are cheap; memory is expensive
• Long-term memory
  • Massive capcacity
  • Can be surprisingly detailed!
  • Aggregate visual experience
  • Provide ‘chunks’ for working memory and ‘guidance’ for attention

Lec.12 Storytelling