Revisiting Data Visualization

October 19, 2020

Data visualization is one of the most interesting data science subdomains for me, and now I decided to revisit my previous progress and build more on top of it.

Actually, I invested a considerable time and effort into data viz. I read books, participated in a couple of Makeover Monday challenges, and even had a visualization blog. So, in this post, I will discuss a bit of my past data viz experiments including projects, tools, and books.

Projects

These are directly taken from my old blog. Head there if you want to look at more detailed descriptions of the projects and my thought process.

Aging America

This one was my first submission for the Makeover Monday challenge. I managed to pull off a customized design to give it a more “infographic” look. One of the most notable ideas are using text color to replace the typical color legends and that lollipop chart in which the points visually form a line.

Cost of a night out around the world

This is more of a fancy-looking Tableau dashboard with interactive functionality. It was another Makeover Monday challenge submission.

Sleep requirement pie chart

I reworked an already built visualization for this project. The original one here was a pie chart. Aside from the fact that pie charts are generally despised, this particular one was meaningless and unsuitable for the data to be presented.

I made it into a visualization that has better aesthetics and is more representative of the data. Not to mention it is customized to be a bit different from the typical chart.

This is actually one of my favorite projects. I was really satisfied with the output.

Tools

Now that we are done with the colorful part, let’s talk about the technicalities. I tried a good number of the major visualization tools, and here are my thoughts about them…

R

Being a statistician who works with R, I did plots in R more than any other tool. R has a ton of visualization functionality that I will put into three categories: base R, ggplot2, and web-based graphics.

• base R: there is a built-in graphics package that low-level with very concise syntax. It is pretty powerful, but like most low-level coding, it kinda gets messy if you take it too far.
• ggplot2: is now the standard go-to package for making graphics in R. It a little more high-level than the base graphics package and has a more precise “grammar” of making plots and adding layers on top of them. Its power is even more realized with the tons of extension packages that open roads to so much stuff built on top of ggplot2.
• web-based graphics: the most common, and the only one I dealt with, is plotly. It is commonly used in shiny apps and has one particularly cool feature which converts any ggplot2-based plot into a web-based plot with just a single function.

No matter the package, R feels like a good middle ground between high-level and low-level. It is code-based and highly extensible, but it doesn’t force you to work with every shape.

Tableau

There is that one time in which I actually left coding and worked with the GUI-based Tableau… What was I thinking?

The first couple of minutes on Tableau make you absolutely amazed! Drag something from here, drop it there, and boom! You got yourself a plot. You can also find a lot of cool stuff people make with Tableau. It is also packed with a lot of sophisticated functionality.

After that, I started making dashboards and got tons of inconsistencies. Some plots would not align well together. Others looked good in their own view but actually looked ugly in a dashboard because everything gets smaller except the text which maintains its size and now looks huge compared to the tiny bar below it. My frustration with these things was way stronger than my will to keep fighting the app or learn to do it in a “smarter” way. Because I believe that the smarter way to do this is with code.

D3

The JavaScript library D3.js works at a lower level than any visualization tool I ever worked with. I learned it about two years ago, but forgot most of what I learned because I didn’t practice. While I was learning it, I was able to make presentable projects like the sleep requirement pie chart rework. D3’s real power lies in its flexibility and interactivity. Here is a link to one of my favorite D3 works made by Nadieh Bremer. I know the design is fundamentally tool-agnostic, but I usually associate D3 with visualizations that look like art.

For data scientists, though, D3 can have a huge barrier of entry. Before getting started with D3, you need to pick up the fundamental web technologies: HTML, CSS, and JavaScript. Also, I noticed that “real-life” D3 work is done with a front-end framework (mostly React).

My future plan

On the theoretical side, I want to re-read some of the popular books that I have already read once, especially Tufte’s “The Visual Display of Quantitative Information” and Knaflic’s “Storytelling with Data”. I might also read new books.

On the practical side, I want to re-learn D3. All these awesome visualizations made with this tool makes it too attractive for me to leave behind. So, it will be my main tool for practice.

I hope the next visualization journey will be fun!