Module 5 - Multivariate Idioms and Data Reduction

Question 1

Multivariate Idioms (V1)

Answer 1

Visualization Idioms for Multivariate data

Question 2

Scatter Plot (V1)

Answer 2

Data: 2 Quantitative Attributes
Mark: Points
Channel: Horizontal + Vertical Position
Tasks: Trends,Outliers,Distribution,Correlation,Clusters
Scalability = 100s of items

Question 3

Scatter Plot Matrix (SPLOM) Scalability (V1)

Answer 3

Color Categories of Data, Reduce Size of Each point, and reduce transparency of each point and make diagonal scatter plots into matrix kernel density plots

Question 4

Parallel Coordinate Plots (PCP) (V1)

Answer 4

Look at values of columns at each row, and show visual relation to other columns

• order of axis/columns matters, as to extract and analyse trends, we need to find an ordering of the attributes that provides us with the most interesting patterns to generate those insights

Scalability = bundle the trajectories (multiple lines, but cannot track individual lines) or blending & transparency

Question 5

Radar Plots (V1)

Answer 5

variant of PCP, but arranged as a polygon.

*better at comparing attributes (due to the shape)

Question 6

Flexible Linked Axis (V1)

Answer 6

each variable is represented as a polygon, connections are represented as line

*very complex

Question 7

Icons/Glyphs (V1)

Answer 7

map multi dimensional data to properties of graphics object

*intuitive and straight forward

Question 8

Why need Data Reduction? (V2)

Answer 8

Screen resolution is limited - cannot represent each individual point on the screen, especially for large datasets

Additionally, complexity (readability and interpretation) is also another issue

Question 9

Data Reduction Techniques (V2 + V3)

Answer 9

1. Filter
   - eliminating elements, display less info.
   * if elements not visible, then doesn’t exist

a. Items
- removing rows, keeping no. attributes same
- tight loop between encoding and interaction (show immediate result of filtering). promotes exploration
-> Scented Widgets

b. Attributes
- removing attributes, no. rows do not change
- use attribute ordering, remove similar/correlated attributes that do not generate new insights

2. Aggregating
   - generating representatives to summaries dataset,
   * details are lost

a. Items
- merge together groups of similar items, represents data/rows with a single mark
-> clustering

b. Attributes
- summaries attributes without changing number of rows
Dimensionality Reduction (preserve meaningful structure of dataset while using fewer attributes)