11
$\begingroup$

In an answer to the question about what characterises the difference between data science and statistics, the main differentiation referred to

Data collection, Data manipulation, Data scale, Data mining and Data communication

The latter being defined as "helping turn "machine-readable" data into "human-readable" information via visualization".

This suggests (although not necessarily implies) that the visualization is located at the end of a data processing pipeline. In contrast to that, the field of Visual Analytics refers to coupling the analytical process with visualizations in order to support reasoning and decision making.

So my question is whether visualizations that support the knowledge discovery and extraction process are considered as an integral part of data science itself? Or to put it that way: Are questions about visualizations and visual analytics appropriate for the Data Science site?

(If this is the case, we should have a and/or tag)

$\endgroup$
1
  • 5
    $\begingroup$ Visualization is definitely an very old and famous approach for extracting information) I would like to see answers and questions that will highlight that aspect, practical solutions and in general. $\endgroup$ Commented May 14, 2014 at 19:08

3 Answers 3

Reset to default
9
$\begingroup$

Yes, I think so, and this site is probably the best one for such questions.

I tend to focus on operational aspects of analytics in work as a "data scientist" -- data cleaning, fusion, large-scale model building, serving, productionization. However I find the popular understanding of "data scientist" refers more to exploratory/investigative analytics -- querying data, building ad-hoc models, and most certainly visualizing data.

A common path I see organizations take through "data science" is:

  • Collect and store data
  • Query data
  • Visualize data
  • Model data
  • Operationalize modeling

In most contexts, the path includes a BI tool like Tableau or Qlikview, connected to the data store. It's often a more basic, useful and prevalent discipline than modeling.

So yes I think this site should welcome well-formed questions on visualization.

$\endgroup$
1
  • $\begingroup$ Do you think it's a nice idea to move this post to the main site. This is a really good qn, and I think more people would benefit from having this in the main site :) $\endgroup$ Commented Jul 25, 2017 at 5:38
7
$\begingroup$

Visualization is often extremely helpful, although it can also (like any metaphorical technique) be used to coerce or obfuscate. So I would not consider it an integral part of data science myself.

As to the site, if I can venture to suggest:

Questions related to how visualizations and visual analytics can help one understand and convey topics in data science should be considered within scope.

But questions about how to use Vizio and OmniGraffle, probably no. ;)

$\endgroup$
2
  • $\begingroup$ Sure, no questions related to someone manually going though a printed-out table of customer data and creating a "visualization" in form of a bar chart that was hand-drawn in MSPaint ;-) I think there already are some visualization techniques that are not at the "end of the pipeline" (e.g. Self-Organizing Maps), and where the interaction of the user serves as an additonal data source (be it only for drilldown and "details on demand"), e.g. for Graph Exploration. So I basically consider questions about these as "on topic". $\endgroup$ Commented Jun 1, 2014 at 18:03
  • $\begingroup$ Great point. Make's me think of the observer phenomenon (quantum physics). $\endgroup$ Commented Jun 2, 2014 at 2:26
5
$\begingroup$

I'd also say yes.

What is visualization used for?

  1. During your own data exploration process, plotting is essential
  2. You have to communicate results to other scientists or peers from your experts group (goal: produce polished scientific graphs)
  3. You have to convince the non-expert stakeholders: the general public, your manager or supervisor, some colleagues - these are different visualization tasks (goal is more like producing infographics)

However, to accomplish this (in my opinion) you also need fluency in different visualizations software ecosystems: your preferred plotting system that you use in your analyis workflow (e.g. ggplot2 from the R ecosystem, matplotlib from python).

For your own communication/publication workflow, you need some HTML/Javascript widget library (such as Shiny, Leaflet, ... or commercial products such as Tableau) to make visualisations interactive, animated; generally to put visualizations online. Often this publication step is optional but I think having this skillset it makes you (as an employee, at least) more valuable.

I think this fluency and awareness of visualization needs of different audiences, and a certain mastery of web-technologies distinguishes a data scientist from a true statistician or ML/AI person who is less skilled in (3).

$\endgroup$
1
  • $\begingroup$ Competence in the points that you mentioned separate the "data scientist" from the pure statistican. The question was asked shortly after the site entered public beta, and in the meantime, there are already many visualization questions. But notably, there are no visual-analytics questions (yet), although the question explicitly referred to that: The case where visualization is not at the end of the pipeline (as in your examples - except for maybe (1)), but instead, is an integral part of the analysis and model building process. Maybe this will receive more attention in the future. $\endgroup$ Commented Jul 21, 2017 at 15:07

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.