10. Visual Analytics

Definition

The science of analytical reasoning faciliated by interactive. visual interfaces.

Visual Analysis

• The procedure in which analysts make reasonings

Visual Analytics

• The system, framework, or discipline of visual analysis.
• Integral approach of combining visualization, human factors, and data analysis.

Early systems

• Cluster and calendar based visualization
• Hierarchical clustering explorer
• XMDV Tool
• GGOBI

Why Interaction?

• Helps analysis to view data more comprehensively
• Helps to correct perceptual errors (in 3d data)

Visual Analytics Models

KDD process

• Waterfall model
• data mining community

  InfoVis Pipeline

• Waterfall again

Views on Visualization

• Explains why wee need interaction

Knowledge Generation Model

---

Implications of Visual Analytics model

• Visual Analytics is an interactive, and iterative process
• They aim to support knowledge generation and decision-making
• Collaboration between analysts and machines

---

What is important then?

Scalability

Shape, threshold, better performance.

→ need for causality. attention.

1. Computational Scalability

• Precomputation
  • 미리 계산해 놓자. 내가 드랍율 미리 계산해 놓은 것 처럼.
  • 미리 aggregation이나 clustering 해도 좋다.
• Query
  • 검색 속도도 빨라야한다.
  • 미리 다 계산해놓으면 빠르겠지만, 메모리 이슈가 있겠지
• Rendering

ex. t-SNE → BH t-sne → GPGPU t-sne

1. Visual scalability

• Overplotting
  • change visual encoding. (opacity)
  • subsampling
  • Aggregation
    • density maps

Observation-Level Interaction

• Manipulate data point and visual patterns directly (DND)
• Require domain knowledge
• Updates parameters to match user intent
  • escape scalability issue

→ Blend parameter with observation-level interactions

• consider parameter as interactive data
• Enable parameter affect visual patterns.
  • Leverage both approaches