Visual analytics is often defined as a highly inter-disciplinary field that brings together disciplines like: visualization, statistics, data mining, human-computer interaction, design … and a bunch of others. That’s fine and this probably why we are very much attracted by it. The question is: how does one become a visual analytics expert? And, how do we teach people to be visual analytics experts?
This simple question was the main topic of a workshop organized at the VAC Consortium Meeting the last week of August. The workshop gathered several leading people in the field, the majority of them teaching in renown academic institution, with the goal of discussing what and how to teach in VA.
I participated to the workshop to tell my story about the infovis course I gave during the last semester and especially to share my personal point of view on what kind of sculpture I think we should make. Here I summarize my ideas and the outcome of our discussions.
What kind of sculpture do we want to make?
One relevant dichotomy we need to address is: do we want to teach people how to build or use VA tools? The problem in my opinion is that we cannot really differentiate between building and using, as is more common in other engineering fields.
Let’s take a VA builder. The data analysis process is so much intertwined with the way a tool works that it is not possible to be a good VA builder without knowing how to analyze data with it. Let’s take a VA analyst then. Well, we have the opposite. Is it realistic to have a VA expert who is not able to build his own tools? I don’t know if this will be possible in the future (some people think we should aim at this, but I am not) but for sure in the current situation there is no way for a VA analyst to skip the hard task of tweaking and building small solutions that fit the problem at hand (for instance can you avoid data crunching with R or Excel?)
What are the unique skills of a VA expert?
In my presentation I was suggesting the following key skills:
- Data Crunching – The ability to manipulate and process data (often dirty, large and scattered across multiple sources)
- Analytical Reasoning / Communication – The ability to understand an analytical problem, to reason around it with other people and communicate the outcome to others.
- Visualization Design and Algorithmics – The ability to extract meaningful patterns out of data and to design visualizations that fit the task.
Pretty tough right? And then we went through a long session of discussions and came up with the “perfect” syllabus.
The “perfect” syllabus.
John (in the photo) was leading the discussion and we ended up with the perfect course we would like to teach. These are the main areas of study we came up with. Note that this list can also be used as a personal guide for people who want to self-study how to become a VA expert.
- Data Analysis Techniques (e.g., Clustering)
- Visualization Techniques (e.g., Parallel Coordinates)
- User Tasks and Issues (e.g., Task Analysis and User-Centered Design)
- Cognitive and Perceptual Issues (e.g., Color Perception, Bias)
- Evaluation and Communication
- Tools (e.g., R, Tableau)
Not too far from what I suggested at the beginning but more detailed.
Reading this list again after few weeks now I must admit it looks quite scary. How are we going to teach all this stuff to one person? Yet, I realize how each of these elements is necessary and how they interact one to another.
- If you are great at visualization design but you struggle with data processing you might end up being overwhelmed or just being left with pretty dull findings.
- If you are not able to focus on the user and build and “appreciation of user tasks and problems” (I loved this sentence from John!) you can easily lose focus and build a useless monster (I fell into this trap so many times! And I see this fault repeated over and over).
- If you start designing a visualization but you don’t know the foundations of perception and cognition, believe me, you are just wandering in the dark. The design space is too big to harness it with your intuition only.
- If you are a great designer, you know every sophisticated data analysis algorithm, and you know the tricks of perception, or in other words you are the perfect geek you are done right? No. You may be great in your cubicle alone but data analysis is about communication and interaction and if you don’t know how to relate with the others and communicate information effectively there’s no job for you.
- If you are good at all these things but you don’t use the right tools or you use some esoteric ones you can easily end up waisting lots of time and/or being isolated.
So yes, it’s tough. But this is what it takes guys.
But wait a minute … what does the market need?
A study plan doesn’t exist in a vacuum, I firmly believe it must be connected to reality. It is not only licit but also necessary to ask ourselves: what kind of job can a VA expert expect to have? This is hard to answer but there’s no doubt we live in the data era and companies will need more people able to analyze their data and communicate their results.
But wait a minute … isn’t that what we call a Data Scientist? Yes and no.
Yes, data scientists are people who need the 3 sexy skills of data geeks and these skills look incredibly similar to the ones I have listed above. Also, the good news is that the demand for DS is increasing steeply and I am totally sure a good VA expert can fill some positions where a DS is requested.
But, as we said above, a VA expert can also be seen as a tool builder, whereas a DS is normally a data analyst (even if a DS might end up building some custom tools). And in this case I am not sure yet how big the market is. I am sure there are thousands of companies out there in need of VA experts to build tools for their needs, but they are hidden in specific niches that are hard to uncover (e.g., pharmaceutical companies, banks). Also there are not many “pure” VA companies out there, but this is going to change soon (e.g., Tableau Software).
What we can do now.
Creating or becoming a VA expert is not an easy task. Also, this is such a new discipline that it is not clear what the place in the market of this profile will be. What can we do about it? I think there are three main areas for improvements:
- Creating VA courses and teaching material – The workshop was a good start. What is really surprising however is the lack of teaching material in form of books and web content. I am surprised how few people tried to organize some of this knowledge somewhere on the web or in a e-book. One middle/long-term goal of FILWD is to fill some of these gaps.
- Creating VA jobs and companies – That’s tough but I am sure it will change in the future. The market is open for a myriad of different kinds of enterprises: big ones, small ones and solo consultants. Also, as I said above data scientists are on the rise and I don’t see a single reason why a VA expert cannot fill one of those positions.
- Connecting the dots – We need people able to talk with everybody and able to talk different kind of languages. The language of computer scientists, the language of data analysts, the language of entrepreneurs, the language of medias and strive to connect all of them and let the ideas circulate. That’s partly the dirty job of bloggers and and we have to make sure we reach out to fill these gaps.
C’mon guys, we have lots of work to do in front of us!!!
What do you think? Is this a realistic picture? Did I miss anything important? Do you have different opinions? Help me make this all thing better by commenting or sending me a message on twitter.
Thanks for reading.