Data Visualization Lecture Notes

Visualization Lecture Notes

• Visualization is used for data exploration and making the unseen visible, leveraging human visual perception.
• Human eyes act as a high-bandwidth channel to the brain, often enabling intuitive understanding via graphical illustrations.
• Data visualization is not simply creating aesthetically pleasing pictures; the goal is to create useful pictures to explore, analyze, and present data.
• Goals of Visualization:
  • Exploration: Use for situations with no prior hypotheses (data exploration).
  • Analysis: Use for hypotheses verification or falsification.
  • Presentation: Communicating existing knowledge about the data.
• Visualization pipeline progresses through transformation, filtering, mapping, projection, and user interaction.
• Humans play a critical role in the visualization process as part of the human-computer interaction loop.
• Data abstraction involves describing data and tasks in generic terms.
• Visual encoding involves selecting appropriate visual representations for data attributes.
• Algorithms concern layout, ordering, rendering techniques.
• Visual encoding design focuses on data types (items, attributes, links, positions, grids), relationships, and spatial representation (tables, networks, geometry).

Attribute Types

• Categorical: Features without intrinsic order (e.g., fruits, colors).
• Ordered: Features with an intrinsic order (e.g., age, temperature).
• Ordinal: Features with an order but discontinuous space (e.g., shirt sizes).
• Quantitative: Features with continuous space (e.g., height, weight).

Data Types

• Static: Data does not change significantly.
• Dynamic: Data changes frequently.
• Qualitative: Data is described in categories.
• Quantitative: Data is described in measurable attributes.

Dataset Availability

• Data is static or dynamic depending on how frequently it changes.
• The user tasks or goals for looking at the data determine the importance and use of the static or dynamic properties of the data.

Data, Tasks and Users

• Quantitative Data: Descriptors of physical dimensions (weight, temperature).
• Ordinal Data: Descriptors of categories with an implied order.
• Nominal Data: Descriptors of categories without an inherent order.

Visualization Channels

• Position: Horizontal, vertical, or both.
• Color: Encoding categorical data, with ordered hues for quantitative dimensions.
• Tilt (Angle): Angle of a visual element.
• Size: Dimensions of a visual element.
• Shape: Geometric form or structure of a visual element.
• Motion: Movement of a visual element.
• Visual encoding analyzes data by combining marks and channels to display data attributes.

Visual Encoding Design

• Data types: Elements, Attributes, Links, Positions, and Grids.
• Data sets: Tables, Networks, Geometry (spatial relations), and Fields (continuous variables).
• Encoding Considerations: Visual channel selection, effectiveness, and expressiveness for the data (focus on channels that are in the data) to communicate only what is present.

Visual Channel Ranking

• Categorical data: Spatial region, color hue, motion, and shape.
• Ordered data: Position (common scale), position (unaligned scale), length, tilt (angle), area, depth, color luminance, color saturation, curvature, and volume.

User Interaction

• User needs, workflows, and limitations should be considered.
• Provide actionable knowledge based on decisions, relevance, and understanding.
• Danger: Misunderstanding user needs can lead to poor visualization output.

Visual Encoding and Interface

• Interface elements: Select visual encodings that communicate only the data characteristics.
• Effectiveness principle: Encode the relevant attributes with the best-ranked channels.

Data Reduction

• Techniques for dimensionality reduction:
  • Linear Combination of Attributes (Linear), Multi-dimensional Scaling (MDS), t-distributed Stochastic Neighbor Embedding (t-SNE), Uniform Manifold Approximation and Projection (UMAP).
  • Filtering/Elimination of elements and attributes
  • Aggregation from data and attributes generation.

Visualization Idioms

• Bar Chart: For displaying categorical and quantitative data. Items and attributes (key-value pairs).
• Line Charts: For quantitative data showing trends over time.
• Scatterplots: For showing relationships between two quantitative variables.
• Histograms: For visualizing distribution of a quantitative variable.
• Boxplots: For comparing distributions of a quantitative variable across categories.
• Violin Plots: Similar to boxplots, but also display the density of the data.
• Maps (Choropleth): For showing spatial data.
• Cartograms: Represent quantitative values by area distortion.
• Dot Maps: Use points to represent data values in a geographical context.
• Density Maps: Display the density of data points in space.

Gestalt Principles

• Proximity: Items close together are perceived as a group.
• Similarity: Items that look alike are perceived as a group.
• Common Region: Items within a common boundary are perceived as a group.
• Good Figure (Prägnanz): Simple and regular figures are easily perceived.
• Closure: Incomplete figures are perceived as complete.
• Continuity: Elements arranged along smooth paths are perceived as a group.
• Figure-Ground: Items are perceived as figures against a background.

Tufte's Principles

• Maximize data-ink ratio: Use as little ink as possible without distorting information.
• Avoid chartjunk: Avoid unnecessary design elements that do not convey information.

Dangers of Depth

• Ranking of planar spatial position is not depth.
• Data position, length, tilt and area are not depth.
• Depth in visualizations is often more complex than in actual 3D.

Time Versus Space

• Visualization idioms to visualize data in time and space contexts.
• Choices to consider for encoding and manipulating time-based data in visualizations.
• Idioms are described by: data, data types, metrics, number of values.

Data Representation

• Data preparation is critical to achieving useful and meaningful visualizations.
• Selecting the appropriate visualization technique, accounting for the characteristics of the data.
• Using the time series data to show trends or patterns over time.

Interaction Principles

• Visual feedback to show the immediacy of user actions.
• Manipulation and highlighting actions.

Multiple Views

• Present different visual encodings of the same data.
• Different perspectives on the same data.
• Show focus and context in a single visualization.
• Spatial region, color hue, motion, and shape are important for visual encoding of spatial information.

Taxonomy

• Categories of user interaction: selecting, exploring, re-configuring, encoding, abstracting and elaborating, filtering, connecting.

Validation

• Algorithm, visual encoding (downstream and upstream)
• Informational usability studies, tasks studies, user experience testing.