3a: Task abstraction

Why do data viz?

Why: An overview

Visualization Analysis and Design, chapter 3, Munzner

Why task abstraction?

Thinking about why in an abstract form, rather than the domain specific way users often talk about data viz.

This will allow us a framework to discuss use cases, which may on surface look different.

Actions / Targets

This proposed taxonomy is from Munzner's framework (Visualization Analysis and Design, chapter 3)

Actions in this case is a verb, and targets are nouns.


Reading: A multi-level typology of abstract visualization tasks, Bremmer, Munzner

Reading: Taxonomy of interactive dynamics for visual analysis, Schneiderman, Heer

Viz designer or user?

Are you consuming the visualization or producing it?

Viz tools fall somewhere along a continuum from specific to general.

On the general side, tools are flexible and allow users many choices what to make.

On the specific side, the tool is curated and choices are limited in how an end user can interact with the data set.

Exploratory Analysis vs Presentation

Storytelling with data, chapter 1, Nussbaumer

3-levels of actions

Visualization Analysis and Design, chapter 3, Munzner

Actions (verbs)

The framework give us a way to think about how and why we do tasks, from the high level (how we analyze) to mid-level (how we search) to the low level (query structure).

Analyze (high level)

Most visualizations are used to consume information in many domain contexts.

Typically it is either for exploratory analysis or for presentation.

When we talk about producing we're talking about generating new data / artefacts - e.g. through annotations and derived columns.

For example, annotating outliers, or tagging clusters in open ended text queries.

Search (mid level)

Target known Target unknown
Location known Lookup Browse
Location unknown Locate Explore

Location: Where does this data exist in?

Target: What do you want to query?

Let's use the carpark example.

  1. What's the utilization of carpark BM29 now?
  2. Which carparks have lot type 'C'? List them.
  3. Which housing estate is carpark BM29 in?
  4. Which housing estate has the highest carpark utilization?

Search Query

If a search queries known targets (lookup or locate), identify returns their characteristics (the column).

If a search queries references based on characteristics (browse or explore), identify returns references (rows).

Query Structure (low level)

Identify — Compare — Summarize is a scale of progression of what to do with the data returned via the query.


E.g. Which carparks have utilization more than 90%?

  • Identify: Just list the data
  • Compare: Perhaps sort by utilization so we can more easily compare
  • Summarize: Group carparks by some other metric, e.g. time, area.

Targets (nouns)

Visualization Analysis and Design, chapter 3, Munzner

What is the target we're returning?

  • Data: are we looking for trends? Outliers? Specific features?
  • Attributes: are we looking for one attribute or many? Comparison, distribution, correlation?
  • Network Data: are we looking at the topology of the network?
  • Spatial Data: are we looking at the shape of a feature? Distribution or clustering of points?

All together: task breakdown

Example: Which carpark is the one with the highest utilization in SG on Monday mornings?

  • Analyze: Exploratory analysis
  • Search: Browse
  • Query: Identify
  • Target: One attribute, utilization

All together: task breakdown 2

Example: SG election. How did the opposition fare in 2020 compared to 2015?

  • Analyze: Presentation
  • Search: Explore
  • Query: Compare, Summarization
  • Target: Trend, Outliers

Thought exercise

How would you frame all these queries?

  • How do I get from SUTD to Sembawang?
  • Which countries in the world have the highest taxes?
  • What is the current PSI like in Singapore? In the region?
  • Is Changi a sustainable district?

Think about how you would frame your final capstone project query.

Industry perspective on why data viz

Sample job Why data viz Tools

Data scientist

Getting insights Python, R. Quick and dirty charts

Business analyst

Insights + simple dashboards for client / management Tableau, PowerBI, Qlik

Data journalist

Storytelling. Presentation to make things clear Graphics / animation tools and libraries

Code artist

Presentation to impress 3D / Animation / sound / interaction libraries

Product frontend

Building a custom product for a specific usecase D3, JS and CSS frameworks

Product backend (e.g. regulator)

Building a custom product for a specific usecase tech architecture, database and warehousing, backend and devops

Viz Tools recapped

How: An overview


We'll cover definitions and sample idioms later in the course.

Questions?

Chi-Loong | V/R