Comment

Comment by Scott Mongeau on March 11, 2017 at 3:38am

Looks like a version of my representation from 2013/14 teaching/consulting/blogging.  Indeed a simplification (Box's "all models are wrong, but some are useful" principal ;-)): 

https://sctr7.com/2014/07/09/twelve-emerging-trends-in-data-analyti...

Some useful additions have been made.  I generally use this in teaching during introductory lectures, or to orient clients with little analytics / data science background or context.

Comment by Daragh Fitzpatrick on February 28, 2017 at 2:33pm

https://www.linkedin.com/pulse/business-intelligence-data-science-f...

Comment by DEREK BELYEA on February 28, 2017 at 9:06am

Managers and decision makers who know little about data science need a place to start their education and this visualization is a start. I agree with Paul McLeod that it fails to tell the whole story but that in my view is not a failure of visualization, as I think his comment suggests.

Systems architects have the same challenge in explaining their craft and illustrating the complex constructs that make up their world. From what I have seen they have developed numerous visual representations that do the job remarkably well.

I would challenge data scientists to find ways to transcend the world of algorithms and make their insights more accessible to the unwashed masses.

Comment by Paul McLeod on February 26, 2017 at 6:02pm

I think this is a very popular oversimplification.
Visuals do allow powerful but biased wetware to engage on decisions, but only to take in a limited number of factors.

To work with large/vast sets of variables, visualization is feeble. Algorithms rule.

Also th focus on understanding, popular as it is, runs as a distraction to building the best engine for making the optimal decision.

I'd add Network analytics, Supervised/Unsupervised, Propensity, Recommendation, Preference Analytics, Identity/Identification Entity Analytics and with Understanding probably add a point for Hypothesis Testing, to differenciate.

Also I'd cover A/B testing, Deployment, Decision Modeling, Optimization/Trade-off, Champion/Challenger and Realtime-scoring.

Surely space on data preparation,munging,feature engineering, discovery,tuning, and information flow are important. And then you get to data engineering and modeling.

All this makes the picture complex and perhaps overly complex, but I believe it is just now overly simplistic and populist. As it stands, this is probably fairly accurately showing the common stereotype of Data Science, rather than its fullness.