The Data Scientist
A new term - well, perhaps not that new - has come up and I’m actually very excited about it. The term is Data Scientist, and since it’s new, it’s fairly undefined. I’ll explain what I think it means, and why I’m excited about it.
In general, I’ve found the term deals at its most basic with analyzing data. Of course, we all do that, and the term itself in that definition is redundant. There is no science that I know of that does not work with analyzing lots of data. But the term seems to refer to more than the common practices of looking at data visually, putting it in a spreadsheet or report, or even using simple coding to examine data sets.
The term Data Scientist (as far as I can make out this early in it’s use) is someone who has a strong understanding of data sources, relevance (statistical and otherwise) and processing methods as well as front-end displays of large sets of complicated data. Some - but not all - Business Intelligence professionals have these skills. In other cases, senior developers, database architects or others fill these needs, but in my experience, many lack the strong mathematical skills needed to make these choices properly.
I’ve divided the knowledge base for someone that would wear this title into three large segments. It remains to be seen if a given Data Scientist would be responsible for knowing all these areas or would specialize. There are pretty high requirements on the math side, specifically in graduate-degree level statistics, but in my experience a company will only have a few of these folks, so they are expected to know quite a bit in each of these areas.
The first area is finding, cleaning and storing the data. In some cases, no cleaning is done prior to storage - it’s just identified and the cleansing is done in a later step. This area is where the professional would be able to tell if a particular data set should be stored in a Relational Database Management System (RDBMS), across a set of key/value pair storage (NoSQL) or in a file system like HDFS (part of the Hadoop landscape) or other methods. Or do you examine the stream of data without storing it in another system at all?
This is an important decision - it’s a foundation choice that deals not only with a lot of expense of purchasing systems or even using Cloud Computing (PaaS, SaaS or IaaS) to source it, but also the skillsets and other resources needed to care and feed the system for a long time. The Data Scientist sets something into motion that will probably outlast his or her career at a company or organization.
Often these choices are made by senior developers, database administrators or architects in a company. But sometimes each of these has a certain bias towards making a decision one way or another. The Data Scientist would examine these choices in light of the data itself, starting perhaps even before the business requirements are created. The business may not even be aware of all the strategic and tactical data sources that they have access to.
Once the decision is made to store the data, the next set of decisions are based around how to process the data. An RDBMS scales well to a certain level, and provides a high degree of ACID compliance as well as offering a well-known set-based language to work with this data. In other cases, scale should be spread among multiple nodes (as in the case of Hadoop landscapes or NoSQL offerings) or even across a Cloud provider like Windows Azure Table Storage. In fact, in many cases - most of the ones I’m dealing with lately - the data should be split among multiple types of processing environments. This is a newer idea. Many data professionals simply pick a methodology (RDBMS with Star Schemas, NoSQL, etc.) and put all data there, regardless of its shape, processing needs and so on.
A Data Scientist is familiar not only with the various processing methods, but how they work, so that they can choose the right one for a given need. This is a huge time commitment, hence the need for a dedicated title like this one.
This is where the need for a Data Scientist is most often already being filled, sometimes with more or less success. The latest Business Intelligence systems are quite good at allowing you to create amazing graphics - but it’s the data behind the graphics that are the most important component of truly effective displays.
This is where the mathematics requirement of the Data Scientist title is the most unforgiving. In fact, someone without a good foundation in statistics is not a good candidate for creating reports. Even a basic level of statistics can be dangerous. Anyone who works in analyzing data will tell you that there are multiple errors possible when data just seems right - and basic statistics bears out that you’re on the right track - that are only solvable when you understanding why the statistical formula works the way it does.
And there are lots of ways of presenting data. Sometimes all you need is a “yes” or “no” answer that can only come after heavy analysis work. In that case, a simple e-mail might be all the reporting you need. In others, complex relationships and multiple components require a deep understanding of the various graphical methods of presenting data. Knowing which kind of chart, color, graphic or shape conveys a particular datum best is essential knowledge for the Data Scientist.
Why I’m excited
I love this area of study. I like math, stats, and computing technologies, but it goes beyond that. I love what data can do - how it can help an organization. I’ve been fortunate enough in my professional career these past two decades to work with lots of folks who perform this role at companies from aerospace to medical firms, from manufacturing to retail.
Interestingly, the size of the company really isn’t germane here. I worked with one very small bio-tech (cryogenics) company that worked deeply with analysis of complex interrelated data.
So watch this space. No, I’m not leaving Azure or distributed computing or Microsoft. In fact, I think I’m perfectly situated to investigate this role further. We have a huge set of tools, from RDBMS to Hadoop to allow me to explore. And I’m happy to share what I learn along the way.