Summary of Placement Strategies

The total number of distinct views are of N-D data on a 2-D data space is shown below.

data-driven (raw):

(1 + D)(N² - N), where D is the number of distortion techniques (e.g., random jitter, relaxation)

data-driven (derived):

$(1+D)\sum_{j=1}^{j \leq k} p_{j}$, where k is the number of distinct derivation algorithms and p_j is the number of variations within algorithm j (e.g., different distance metrics).

structure-driven (linear ordered):

$N \times FP (1 + SP + OL)$, where FP is the number of filling patterns, SP is the number of methods which introduce white space for separation (space-padding), and OL is the number of methods which distort the placement to allow overlaps.

How to select? Factors include:

• Characteristics of data (size, distribution)
• the purpose of the visualization (presentation, confirmation, exploration)
• the specific task(s) at hand (e.g., detection, classification, or measurement of patterns or outliers).
• the skills of the prospective user of the visualization.
• domain knowledge that can lead to "intuitive" mapping
• trade off between efficient screen utilization, the degree of occlusion, and distortion of the values being used to position the glyph
• whether to impose an implicit structure (what comparison metric to use?)