Database Reference
In-Depth Information
In explaining his thesis that “most people's charts and graphics look like crap,” com-
mentator Ross Crooks makes the seemingly-obvious but often ignored point that “good
data visualization relies on good design. And it's about more than just choosing the right
chart type. It's about presenting information in a way that is easy to understand and intuitive
to navigate, making the viewer do as little legwork as possible. Of course, not all designers
are data visualization experts, which is why much of the visual content we see is, well, less
than stellar.”
The fundamental rules are so full of common sense they barely need to be enunciated:
Don't use complex animations when a simple pie-chart will do the trick. Don't arrange
data non-intutively. Don't design in a way which obscures relevant data.
Never
use 3-D an-
imations, which distort perception and thus can skew the viewer's understanding of data.
Assure data is presented and understood accurately by sticking to 2-D shapes. Finally,
nev-
er
go for what “looks good” over what is clean, precise, accurate, and easily-digested.
Writing in his now-classic book
The Visual Display of Quantitative Information
, Ed-
ward Tufte tells us: “Graphical excellence is that which gives to the viewer the greatest
number of ideas in the shortest time with the least ink in the smallest space.” To this he
adds a salient warning. “Cosmetic decoration, which frequently distorts the data, will nev-
er salvage an underlying lack of content.”
In sum, the role of data visualization in Data Science is two-fold.
Early in the process, well-designed visualizations help the Data Scientist discern and
discover new trends and patterns in data which in turn yield hints for further research and/
or indicate actionable BI results. With this tool the Data Scientist makes initial discoveries.
The Data Scientist and his team are the consumers and users of the visualizations at this
stage of the process.
At the conclusion of the process, visualizations must be created, customized, and
geared toward explaining results to the ultimate consumers of the research - people beyond
the Data Science team. This is where concise simplicity is most needed: telling the story
of the research and results cogently and efficiently, with a minimum of that odious thing
called
noise
.
