Visualizing Categorical Data with a Reorderable Matrix

1
Visualizing Categorical Data with a Reorderable
Matrix

• By Darya Filippova

2
Categorical Data

• Binary male, female. 
• Maps to 0,1
•  
• Ordinal cold, tepid, warm, hot. 
• Can map to 1, 2, 3, 4, but no mean (what is
  2.5?)?
•  
• Nominal sedan, hatchback, van, truck.
• Cant map to numbers.

3
Visualizing categorical data

• Histograms (1D)?
•  
• Trellis (2D)?
•  
• Mosaic plots (2D)?
•  
• Matrix (2D)?

4
Which one is more important?
5
(No Transcript)
6
Mosaic plots
7
Matrix
8
Reorderable Matrix

• Makinen, Siirtola, 2000
• Based on Sugiyama algorithm for graph layout
  minimizing the of crossovers in corresponding
  graph results in a clustered matrix
• Works especially well for categorical variables
  of high degree

9
Algorithm
10
Large categorical datasets

• Road accidents
• Genome data
• Taxonomy of living organisms
• Binned continuous variables
• How do we analyze these systematically?

11
Rank-by-feature principles

• Seo, Shneiderman, 2002.
• Study 1D
• Study 2D
• Use ranking to guide user's insight
• 3D and further

12
Demo
13
1D Rankings

• The most variance
•  
• Normal distribution fitness
•  
• Outlier count by Inter Quartile Range (IQR)?
•  
• Unique count
•  
• Uniformity (entropy)?

14
2D Rankings

• Most variance
• Crossover
• Percentage filled
• Percentage empty space
• Outlier count by IQR
• Uniformity (entropy)?

15
Questions
?