The Data Analytics Handbook2 PDF
The Data Analytics Handbook2 PDF
T R I S TA N TA O Content
Tristan holds dual degrees in Computer Science and Statistics from UC Berkeley (14). He first began working as a quantitative technical data analyst at Starmine (Thomson Reuters). From there he worked as a software engineer at Splunk. He has experience working with various Machine Learning models, NLP, Hadoop/Hive, Storm, R, Python and Java.
E L I Z A B E T H L I N Design
Elizabeth is in her third year studying Computer Science at Cal with a focus in design. She is President of Innovative Design, a student group of visual designers and photographers, and Layout Director for BARE Magazine, Cals fashion and arts publication. She has interned at LinkedIn, TechValidate, and UC Berkeleys EECS Department. She will be a Product Design Intern at Khan Academy this summer.
hat exactly do the sexy Data Scientists do? We start with this simple question. What other professions are
there in Big Data? What tools do they use to accomplish their tasks? How can I enter the industry if I dont have a Ph.D. in Statistics? The genesis of The Data Analytics Handbook stems from our own internal frustrations with these questions; as recent graduates of UC Berkeley (Go Bears!) in statistics, we saw a burgeoning industry craving curious minds but disconnected from its potential employees because of a lack of understanding of what the Data Analytics industry is and of what it requires. And so, we set out to discover the answers for ourselves by reaching out to industry leaders, academics, and professionals. This handbook is the first of three parts and will focus on the experiences of current data analysts and data scientists. What we discovered in our research is that while each interviewees response was highly informative, their knowledge was fundamentally incomplete; the truth is that the Big Data industry is still so nascent that there is no singular definition of a data scientist or data analyst. In fact, the two terms are often used interchangeably. While we continue to ponder the big questions surrounding Big Data, we believe that our research has shone light in an area unexplored, making our interviews that much more valuable to you. The firsthand experiences of those in the trenches, those leading the troops, and those crazy Ph.Ds, have not been shared until now. We hope you enjoy reading the interviews as much as we enjoyed conducting them!
T O P 5 TA K E A W AY S
FROM CEOS AND MANAGERS 1. The Information Revolution is real.
Data analytics will be applied everywhere but it will not surface itself to the end user like most good technologies. Data oriented decisions is a reality for all decision making in the future.
5. The most important skill is being able to ask the right questions.
The power of data analytics is in taking open response questions and framing them to be multiple choice. Therefore if you have the ability to filter a million questions into options A through D, you are a data scientist for hire.
TA B L E O F C O N T E N T S 05 D E R E K S T E E R
M O D E A N A LY T I C S
10 14 18 22 26
DEAN ABBOT T
SMARTER REMARKETER, INC.
TOM WHEELER
C LO U D E R A
MIKE OLSON
C LO U D E R A
ROHAN DEUSKAR
STYLITICS
30 G R E G L A M P
Y H AT
34 A L I S Y E D
PERSONTYLE
38 D AV E G E R S T E R
BIGML
DEREK STEER
C E O + C O - F O U N D E R AT M O D E A N A LY T I C S
D E R E K is the CEO and co-founder of Mode. Previously, he worked on the Analytics teams at Yammer and Facebook, both with a focus on monetization. M O D E A N A LY T I C S is building tools that eliminate the costs of sharing and discovery to make analysts and data scientists more productive.
05
Can we go into your background rst? You have a unique perspective in that you have both the experience of a data analyst at Yammer and the experience as CEO of Mode Analytics.
I graduated from Occidental College where I majored in economics. My brain is just wired for economics I think of everything in terms of opportunity cost and tradeoffs. I wanted to take a job where I could apply economic thinking, so I ended up working for an economic consulting firm doing analysis to support expert testimony in antitrust cases and mergers. I worked there for a few years, then made my way to Facebook. In consulting, managers tell you the beginning and the end of the problem youre solving, and its up to you to figure out the middle part. At Facebook, I spent a lot more time thinking about the beginning (how to frame a problem) and the end (what the output/ recommendation should look like). I wanted to be somewhere smaller, though, so I joined Yammer as one of the early members of its Analytics team. At Yammer, I worked almost exclusively on sales, marketing, and customersuccess related projects a natural fit after working on monetization at Facebook. Yammer was similar in that they sought to democratize data. A part of my job was to enable my internal customers to make better decisions using data without necessarily having to ask me for recommendations every time. I did this by creating lightweight tools either web apps or reusable SQL queries that my customers could run on their own. When a particular request required ad-hoc analysis, I spent most of my time on the beginning and ends of the problem, and almost no time executing. That is to say, I thought of clever ways to solve problems by counting and summing things rather than building complex models. Model building takes a lot more time and doesnt necessarily lead to better business decisions. My former boss, Peter Fishman, deserves a lot of credit for enforcing this type of behavior on the team, a behavior that, I believe, was critical in making the Analytics team at Yammer successful. I left Yammer after two and a half years or so along with a couple of teammates there to start Mode.
What are the most important skills you think a data analyst needs?
Creative thinking and the ability to solve problems. I hate that answer though because its so vague. Its true, but its vague. I will try to give a concrete answer.
06
I think it makes sense to explain this in the context of an interview question, because Im specifically trying to measure problem solving ability. The way I assess creative problem solving in an interview is to ask about a specific project the analysts completed at a previous job and request that he walk me through the tradeoffs he faced and how he determined the correct path. When I do this, Im simply listening for sound logic. If anything sounds odd, Ill probe deeper. If Im not an expert in the type of business problem the candidate is solving, he should be able to easily answer all my questions about how to perform the best analysis. I also use some hypothetical questions. I like to ask people about pricing because its very nuanced. What should we charge for Yammer? It requires some fundamental knowledge of Yammers business model, so its also a quick check to see if youve done your research about how Yammer works. Assuming you have, its still a multi-faceted problem. Good candidates are occasionally able to talk through the complex work they might do to find the right price, but what I really love is when people break down problems into the simplest elements. The most important thing about setting price is understanding what you are trying to optimize: Youre optimizing revenue, which is price times quantity. Its a simple equation but even just saying that gives you an understanding of the levers you have to pull and what theyre going to affect. If you can nail that part, many other parts of the problem become clearer.
What are aspects of the job as a data analyst that you enjoyed and what are aspects that you didnt enjoy?
What I have enjoyed the most is working with great people. I would do any job as long as I liked the people I was working with. Within the world of data analysis, I might be a little weird in that I dont particularly enjoy learning new technical skills. Its something I do because I have to, but what I really enjoy is the feeling of solving a new problem that leads to some sort of action. If other people are applying my work and seeing returns from it, that makes me happy.
How pervasive do you see the demand for data analytics becoming? Do you see the demand going past the metropolitan hubs of SF/NY?
I have a friend who lives in Arizona and works for CVS where they have hundreds of people who do data analysis. Its already there. Ill give you
07
another example. SAS has 70,000 customer sites in the US. Its not just a NY/ SF thing. The question is whether people are doing it well. My suspicion is that the data analysis that people are doing across the US is actually relatively similar in nature. Even the data looks very similar from company to company within a given vertical. Theres a huge opportunity to bring together all of these companies, as well as individuals who are doing analysis in a non-professional setting. Thats part of what were trying to do at Mode.
What are some of the softer skills required for a data analyst?
To some degree, analysis or data science is all about trust. You can deliver the best insights but if your customers dont trust you, its useless. Analysis that does not turn into a decision is basically a waste of time. Establishing that trust is a tricky thing to do and requires a fair degree of interpersonal skill. Its partially about your customers believing that youre smart and capable enough to actually deliver accurate insights. Once youve got that down, its about communicating ideas clearly. I love linear regression. Its fantastic because its very easy to explain to a non-technical person what a coefficient means. You can identify which variables make a difference, and whether they are positively or negatively correlated with the outcome you want to affect. You dont even need to show your customer the regression output you can just explain it in plain English. More complex models make this a bit tougher. And, of course, counting and summation are the easiest things to explain.
What advice would you give a non-technical person trying to gain more technical skill?
08
Its pretty easy to learn new technical skills. Or, at least, it can be. I dont really believe in learning or applying skills simply for the sake of doing hard stuff. I actually think its dangerous, because thats the type of thinking that can lead you down rabbit holes when youre doing something important thats time-sensitive. I think that the best way to learn skills so that you can apply them practically in the future is to start with a project, then learn all the skills necessary to complete it as you go. If you cant think of a project that incorporates the skill youre trying to learn, you should probably ask yourself why youre even bothering to learn it in the first place.
09
DEAN ABBOT T
C O - F O U N D E R AT S M A R T E R R E M A R K E T E R , I N C .
D E A N holds a B.S. in computational mathematics from Rensselaer Polytechnic Institute and an M.A.M. in Applied Mathematics from University of Virginia. In his early career, he worked as a research scientist at companies including Lockheed Martin and Elder Research, Inc. He also founded Abbott Analytics. He is now the co-founder and chief data scientist at Smarter Remarketer, Inc. He also recently published a book: Applied Predictive Analytics: Principles and Techniques for the Professional Data Analyst [Wiley]. SMARTER R E M A R K E T E R collects comprehensive behavioral data. This
includes traditional web analytics style data, but also a compliment of specialized data required for behavioral marketing best practices. Through Smarter Remarketers Metrics feature, the company shows retailers exactly how people are behaving, why they are abandoning and how much revenue can be recaptured.
10
How would you dene the role of a data analyst vs. a data scientist?
Data Science is a relatively new way to describe analytics. Data analysts are more business focused, likely to be less academic focused, have a bachelors degree, and are trying the solve business questions. Data Scientists are more dominated by advanced degrees holders, more likely to work with big data tools (such as NoSQL, Hadoop, MongoDB, and Cassandra), utilize R/Python, and also more likely to be serious programmers. The data science position may be a superset of most analyst roles, whereas data analysis may be a subset of data science.
Would you say that data scientists and data analysts work in disparate groups?
I think so. The data scientists I work with are more research oriented. They are more cutting-edge and are working to create new ideas and new algorithms. They also try to solve problems where the traditional databases dont work well. Data analysts are more likely to work within a business group. In certain situations where there is a specific data science group, people are more likely to be labeling themselves data scientists.
How deep of statistical understanding / programming skills do you need to become a data analyst?
More math is always better, especially if youre doing cutting edge work. Nonetheless algorithms such as Decision Trees, Neural Networks, and Support Vector Machines will work right out the box even if you dont know how they work mathematically. I have a book coming out called Applied Predictive Analytics, which teaches people predictive analytics while using minimal
11
mathematics (some algebra and basic stats terminology). It is important to know when/how to turn the nobs in statistical models. Ultimately, math isnt the most important metric for an analysts quality. But rather love for data and an innate curiosity make for a great analyst; I call it the freakonomics mindset. Also, knowing the details/strength/weakness of the algorithm is always helpful. This is especially important when the model doesnt work: why is the decision tree not working? Can the data be skewed? You learn to ask the right questions.
How do you develop the skills required to be an effective data analyst? Your previous work involved your client bringing in an outside expert (you!), but how can an undergrad/recent-grad develop the same skills?
Two ways. First, take a few classes to get some fundamental understanding of data. Not necessarily databases, but rather about data and what data means. Basic stats and basic courses in machine learning can help (know thy data!) In university classes you learn theoretical details and implementations etc. But the details are not required and are no longer used once youre out working in the industry; university educations dont prepare you for business applications. Some schools with application-based courses (UCSD, UCI) have after-hours programs for business people, which are helpful. Second, start building models. Work on projects. It helps to work with someone who has done it before. Data preparation is harder to teach because there are so many ways for you to do it incorrectly. It is hard to teach in a way where you cover all incorrect approaches. It is a bootstrapping method; try co-op, internships, working in a company etc.
How do I make a switch into data analytics after Ive been working in the industry?
Take online courses. Read some books (Data Mining Techniques Linoff, or my Applied Predictive Analytics book). There are books that will teach you applications, good principles, and how to approach the problems. Once you understand the process better, get on Kaggle! Handbook of Statistical Analysis and Data Mining applications Elder, Miner is also a good read.
Which is more important for a data analyst: intellectual curiosity/intuition V.S. deep statistical knowledge?
12
The curiosity. The understanding of statistics can be gained, but the curiosity is more innate; if youre not naturally into working with data, you can be a fantastic analyst, but you become a specific type of analyst. These analysts may not care enough to build a better model. They lack creativity and the determination to go the extra mile. Theyll need handholding and will struggle to resolve new problems. With curiosity about the data and an interest in why the data looks the way it looks, you can overcome a lack of mathematics training. I once worked with an analyst who had only high school math but became a great modeler. In the beginning, he didnt even know algebra well and certainly didnt know any calculus or linear algebra. But he had good intuition and determination, and ultimately became an excellent analyst and modeler.
13
TOM WHEELER
S E N I O R C U R R I C U L U M D E V E L O P E R AT C L O U D E R A
T O M spent most of his career as a software engineer working on the design and implementation of software for major financial, healthcare, and aerospace firms. He started using Apache Hadoop in 2008 after he recognized that it solved a common problem hed observed at several companies; namely, how to store and process ever-increasing amounts of diverse data at reasonable cost. In 2011, he joined the curriculum development team at Cloudera so that he could share what hed learned with others. C L O U D E R A is revolutionizing enterprise data management by offering the first unified Platform for Big Data: The Enterprise Data Hub. Cloudera offers enterprises one place to store, process and analyze all their data, empowering them to extend the value of existing investments while enabling fundamental new ways to derive value from their data.
14
What are the most valuable skills an aspiring data analyst should develop?
The most important skills are understanding which factors help a business succeed and determining the key metrics by which those can be evaluated. These skills ultimately provide the foundation needed for constant experimentation and iterative improvements that will help uncover important patterns and help a business thrive.
Our understanding is that Data Analysts require a certain level of technical skill and experience; do you not feel that is the case?
While Data Analysts do require some amount of technical skill, its different than what a programmer might need. Instead of developing an e-commerce application in Java as a programmer would, the data analyst might write queries in a language like SQL to analyze the profit margins or sales trends for that Web site. Even though these skillsets are a bit different from one another, both require specialized training. Let me also contrast this with the role of a Data Scientist. Given a set of information like the order data for an e-commerce site, the Data Analyst knows how to answer important questions like, how many products did we sell this year compared to last year or which product categories have the highest profit margin? The Data Scientist, on the other hand, knows how to gather new data as well as query it. They approach the problem scientifically, designing experiments, forming hypotheses, and building systems to collect the data needed to validate the results. In other words, the Data Scientist knows how to ask questions as well as answer them, and has the software engineering skills needed to build new tools when needed. Thats a rare combination of skills, but one thats increasingly in demand as businesses become more datadriven.
Can you expand on what you mean by a Data Analyst being able to ask the right questions?
Data analysts need to understand the constraints of a business and what makes it successful. For example, if youre working for an airline, then youd want to know which things affect costs, which things affect revenue, and which things help you to attract and retain customers. Fuel is a major cost for an airline,
15
so you might do some analysis to project the future cost and buy in quantity when the price is low. In terms of increasing revenue, consider that customers are typically more sensitive to price when buying tickets for personal travel than for business, since the fare comes out of their own pockets. You might therefore attract new customers by offering low-cost fares to popular vacation destinations like Florida or Hawaii and offering higher-cost fares with perks that appeal to business travelers going to destinations like New York City.
Do you think there is a less experienced version of a data scientist that an undergraduate degreed person could achieve?
Certainly. Not all data scientists have a Ph.D. They tend to have a mix of skills in statistics, software development, research, business, and so on. Thats a really rare combination and there are currently very few PhD programs that give you exposure to all of them. Just like there are lots of amazing programmers who dont have a PhD in computer science, so too are there amazing data scientists who started working after getting a Masters or Bachelors degree in one of those areas. If they have an inquisitive personality and a lot of self-motivation, they tend to can quickly gain any other skills they need through real-world experience.
16
and getting certified is a good way for people to demonstrate their expertise and stand out in a competitive job market. Finally, self-education is a great option for learning new skills. Udacity recently partnered with Cloudera to offer an Intro to Hadoop and MapReduce course, for example, and theyve got some others that are really relevant to people just getting started in analytics, such as Intro to Statistics, Intro to Data Science, and a range of machine learning courses.
How clearly do you see the role of the data analyst as being dened in todays job market? Do you see it becoming a more standardized position?
Data Analyst is one of those job titles that mean slightly different things for each industry or organization. Its kind of a general term like programmer; for example, some programmers design new programs while others do maintenance on existing ones. Some are doing low-level things like writing device drivers and others are doing high-level things like developing Web applications. Although I dont really see the Data Analyst role becoming well-defined across all industries, I do see demand for that role increasing across all industries. To paraphrase something that Googles Chief Economist Hal Varian said in an interview for the Freakonomics blog in 2008, the key to success is being able to offer value around some commodity thats becoming ubiquitous and cheap. Organizations are producing exponentially more data than they did even ten years ago and this trend will continue far into the future. The people who have the skills needed to extract value from all of this data are going to be in very high demand.
17
MIKE OLSON
C O - F O U N D E R AT C L O U D E R A
M I K E was the founder of Cloudera and is currently the Board Chairman and Chief Strategy Officer. He graduated from UC Berkeley with a Bachelors of Arts in Computer Science and a Masters of Science in Electrical Engineering and Computer Science. C L O U D E R A is revolutionizing enterprise data management by offering the first unified Platform for Big Data: The Enterprise Data Hub. Cloudera offers enterprises one place to store, process and analyze all their data, empowering them to extend the value of existing investments while enabling fundamental new ways to derive value from their data.
18
How would you describe the value that data analysts bring to a company? How about data scientists?
Data is newly available in volumes and in variety previously impossible to capture. That explosion in growth is a big opportunity companies that figure out how to process and analyze it will rack up real competitive advantages over those that dont. Every organization needs to become data driven. Fortunately the tools available have evolved quickly right alongside the data. Its now easy and inexpensive to store, process and analyze that data. The skills required to do so are still reasonably thin on the ground, and the profession of data scientist has emerged to fill that gap. Its a rich area for innovative smart young folks to get into. In the long term, the data science expertise thats between the ears of people needs to be rendered into software, so that we can scale out. Its hard to make huge numbers of really smart new workers; its easier to capture smarts in applications and tools and make them available to traditional businesspeople who can use them to explore their markets, evaluate their competitors and engage with their customers better. So I believe the best data scientists will also be competent software developers.
Cloudera sells its big data infrastructure to banks, hospitals, etc. but who takes care of the analysis? How much analytic consulting does Cloudera have to provide?
About two thirds of our revenue today comes from software subscriptions, and we see that fraction growing over time. Our annual renewals model means that most of our installed base renews every year, and the work we do to find new customers pays us off for years to come. Of the remaining third, the money is split about equally between training and consulting. We view both of those as ways to drive our subscription numbers up we make customers smarter with training, and we help them build analytics with consulting, and theyll consume more of our software. We have a rich ecosystem of systems integration and consulting partners including big names like TCS, Accenture and Cap Gemini, and smaller firms like Think Big and Mu Sigma. They do a lot of work with our customers. One financial analyst estimates that for every dollar spent on software licenses in big data, fourteen dollars are spent on hardware and integration services.
19
Overall Cloudera seems to answer the question of big data capture and storage. What do you see as the answer for the question of analysis? Put another way there is a clear talent shortage in data analytic skill, how do see this gap being lled?
There is a lot of innovation in that space right now. Tons of startups are attacking different vertical-specific analytic opportunities. Others are building horizontal capabilities machine learning, for example, is useful across most of our installed base. As I said before, those techniques need to get rendered into software and made consumable by traditional enterprises. Business people are plenty smart enough to use the tools and interpret the results, they just need help from data scientists who build the apps that do machine learning by way of random forests and present the results in useful ways. When relational databases were brand new, no one knew SQL, there were no apps and business people couldnt use them very easily. That changed with the huge economic opportunity that the database market created for app and tool developers. Same thing is happening today in big data. The landscape is way, way better than it was when we started six years ago!
Currently it seems like big data technology is really only necessary or possible with the largest technology companies or corporations but do you see big data applications (using Hadoop/MapReduce) for small businesses eventually? If so, how?
I think its already there. Most of the public cloud adoption of Apache Hadoop is by small and medium-sized enterprises. I also think that people will consume machine learning as a service, bundled into business apps they require. Think about how you use Amazon and Google. Theres tons of analytics and personalization behind the scenes there. Youre getting Hadooped all the time. You may not run that software yourself but you are using it extensively by way of the online services you consume. More cloud apps will mean more big data analytics, made available in useful ways to regular people and small companies.
What are the limitations of Big Data? What are the pessimists of the information revolution arguing?
20
Clearly privacy is a big deal, and we have work to do there. Technology determines what can be collected; we must formulate a policy and ethics that tell us what is okay to do with it, and we need a governance framework that can impose meaningful penalties on misuse. Plenty of opportunity but the need here is pretty urgent. The problem is worse because of inconsistent laws among countries. The EU has a pretty comprehensive data policy, with each country having adopted a consistent framework at the national level. The US has essentially no Federallevel data policy. We spell out policy in some key verticals HIPAA, FCRA but leave it to states, and worse to companies, to craft and enforce policies generally. That patchwork makes it very hard to do business. The internet doesnt really know about geopolitical borders. When EU data privacy and safe harbor provisions impose rules on data that crosses borders, it creates enormous friction for businesses and NGOs. Friction equals cost.
What advice do you have for undergraduates or young professionals aspiring to enter the industry?
My advice to young people pursuing any course of study is the same: Find problems that really interest you, really blow your hair back, and find a way to work on them. Chase interesting. Dont chase money. Money will come if you love what you do so much that you become a world-class expert and plow huge effort and time into your work.
21
ROHAN DEUSKAR
C E O + C O - F O U N D E R AT S T Y L I T I C S
R O H A N is the CEO & Co-Founder of Stylitics. His goal is to help consumers and retailers leverage the data in consumer closets to power smarter decisionmaking, better recommendations, and in general, bring the fashion industry into the digital age. Rohan was selected as a 2014 New York City Fashion Fellow and has spoken on the intersection of fashion and data at events like NY Fashion Week and DataBeat. He has an MBA from the Wharton School. S T Y L I T I C S is a fashion analytics company that leverages a proprietary stream of consumer closet data to help retailers improve CRM, recommendations, and targeting. Stylitics has built the worlds largest consumer clothing panel using its popular digital closet platform.
22
How would you dene the role of a data analyst? How would you dene the role of a data scientist?
As we see it, a data analyst is somebody who can sit down with a set of data, in some simple form like CSV, and is able to do two things. Firstly, understand the patterns and trends in the data. Secondly, be able to manipulate the data through various tools in order to understand the fundamentals of the dataset. That could be knowing about data that is missing, knowing additional data that is needed, or some other kind of insight about the characteristics of the data were working with. Basic tools like pivot tables are a requirement, but a good analyst is somebody who can come back to me and say Im worried because Ive noticed that average price is too low for this product category. It tells me the data is weak here. Or maybe were counting blank entries as zeroes. The best data analysts act like detectives.
Given trends in technology are you requiring higher technical skills for your data analysts with languages such as R or SQL?
Knowing SQL is an advantage because it brings you one level closer to the database, allowing you to craft your questions in a way you are unable to when using pivot tables. Given that, I would rather find somebody who doesnt know SQL but has a really good eye for finding patterns within the data, than find someone who is technically proficient but doesnt have the right attitude or mindset towards itthat innate curiosity.
Our approach has always been different because we have a unique data set: post-purchase behavior in fashion, provided by consumers themselves. This is hugely valuable for retailers and has never been available before at scale. Our approach to data is to make it accessible and relevant for our end customer. We think of things from the customers point of view. What is the day-to-day of the fashion business? What is the data we have that relates to their business needs? Can we answer their questions better or faster than they can using the data that we have? We want to give them data they can pull up on demand in the 5 minutes before they go into a meeting. We also try to be very disciplined on data cleanliness. There have been many chances to get more consumer-generated data with less cleanliness. For example, on our mobile apps we could let people simply write in the product name in a free-form text field versus making them select from dropdowns. It would make adding items easier and generate more data. But then wed have much less ability to easily sort and categorize the data and provide useful insights. Other companies have opted to go with letting users tag photos in an unstructured way...as a result, our data set is marketable whereas theirs is not.
24
of metrics religiously and base their decisions only on data. Were helping companies to use a combination of data and human judgment, which is where I think the real success happens.
What kind of interview questions would you ask to hire a data analyst?
We are actually in the process of hiring a couple analysts right now. First of all I want to know, have they worked with large data sets before and to what extent? The next thing I would do is show them a data set, for example all the details for the clothing purchases made by everybody in your university, and ask them what are the some of the insights you could get from that? I want to know how they structure the questions about data themselves, without prompting from me. e.g. I know you asked only for the impact of weather on category sales, but I decided to also check if shades of color sales were affected by weather and found something interesting... We will also give them a raw data set to take home and have them share five interesting things they see in the data. They would also be asked to present their findings in a couple of PowerPoint slides because part of the data analyst role according to me is being able to convey your findings to people who havent spent the time you have on the data.
What are your thoughts the development of technology outpacing the demand for data analysts?
Yes, in a sense our technology removes some of the need for data collection and analysis by humans. Thats part of a macro trend. But with more companies focusing on personalization, targeting, CRM, and other data-driven approaches, the overall need for data-savvy people will grow as well. The existence of data mining and analysis technologies shouldnt dissuade students. For example, spell check exists, but you still need to know how to spell words and use them in the right context Its important for everyone to know how to think about data, how to frame the right questions, and how to work with large datasets. All of that will serve you well regardless of the specific tools that are being used.
25
26
What led you to work in data analytics and why are you passionate about it?
For me it goes back to when I was an MBA student and took a marketing research course. At the time conjoint analysis, a form of tradeoff analysis, was gaining popularity. We gauged how much customers will pay for different features and how much they value one product attribute against the other. In doing so I saw the value of analysis when it came to making good marketing decisions. I realized I was more interested in the research side of marketing rather than the tactical side so I ended up doing my PhD and getting a lot more training in sophisticated analytical methods. Since then I have worked on the analytical side of marketing.
What aspects of your job at Flurry do you enjoy, and what aspects do you not enjoy?
Flurry has big data, and I enjoy seeing what I can do with that much data in terms of really honing in on differences that you might not be able to see with self-reported survey data. On the downside, the data management required when you have that much data is a bit annoying. I come from more of a survey research background where the data is much more clean and more structured. With our data there is a lot more work that goes into pulling the data you want off of our system. Sometimes thats even impossible or requires an engineer. Sometimes the file that you generate is too big to open in a single computer. So before you do actual analysis, you have to do a lot of work to get it ready and into the format you need before you get to that aha moment.
Why should a recent grad work in this eld, as opposed to a more traditional job?
Im actually not sure they have to choose and I would almost encourage them not to choose. I believe what will be increasingly valuable in the future is
27
not what is valuable now, but rather a combination of quantitative skills plus domain-specific skills (such as marketing). There probably will be more jobs for people with that combination of skills than for people who are in pure data science. Your data comes from a specific context; the context gives you clues as to what your data might be trying to tell you. The quantitative tools are there to help you better understand the context.
How do you think the role of a data analyst has changed, since it has become popular, and how do you think it will evolve?
I think there will more emphasis on understanding statistics. Analysts will have to be able to tell you the best technique to extract answers, given a specific data set. People will have to know about the different assumptions in regression; what different distributions looks like and when to use different clustering algorithms. That kind of statistical knowledge will help you when youre trying something and it isnt working. It helps you diagnose the problem. Many data science people now have a more solid understanding of programming than statistics. Many people know R and Python and know how to fit a clustering algorithm to a data set, but lack much knowledge of the underlying statistics that they are applying so are almost using them as a black box.
How do you think your company uses data analytics as a comparative advantage to other companies?
One of our products is analytics, so in that sense we help provide a comparative advantage to other companies by helping them understand their users. Beyond that we are also taking the data we generate through our analytics product and using it to power advertising. The better our analytics are, the better we are going to be at showing ads that are relevant and interesting to you and therefore have a good return for advertisers.
How soon do you think skills related to Big Data (i.e. Hadoop/ Hive/Pig) will be a necessary skill for data analysts?
It would awesome to have those skills now. I dont know MapReduce and Hadoop and it would be really useful if I did because I would be much more selfsufficient in terms of being able to go and get the data I need rather than being dependent on someone else to get it for me. Over time, I suspect that tools like Pig/Hive will get developed (and ones that are even more user friendly), that will let you query even very large non-relational databases without knowing the
28
programming concepts behind them. The amount of data that we have and the uses for it are such that we cant be dependent on the small number of people who have high-level programming skills. Also, if youve invested enough time in your career to have that level of software expertise you may not have subject matter and statistical expertise. What it gets down to at the moment is having a team of people who have these different skillsets. Over time, the tools will catch up and you wont have to know how to program to get the data.
What do you look for in a data analyst, and what specic skills or experience do you favor? How do you interview them?
They would need to know in which context to apply what kind of tools. Do they know the context for factor analysis, or when regression is appropriate? The next step, now that you know which tools to apply when, is how do you deal with challenges such as your data is incomplete, or where you have results and they havent come out in a way that you expected. What is your process then? Do you have a process for checking for bad data? Finally I check for the ability to translate results from data to a solid explanation of what does that mean?
29
GREG LAMP
C O - F O U N D E R + C T O AT Y H AT
G R E G graduated from University of Virginia with a B.S. in Systems Engineering as well as a B.A. in financial mathematics. He worked as a senior data analyst at comScore and also as a product manger at On Deck Capital. He is now the co-founder and CTO of Yhat. Y H AT is a cloud solution for deploying predictive analytics at your company. Data scientists create pipelines to ingest and clean data and train and deploy models. Models are instantly accessible via REST, low latency streaming API, and batch mode.
30
Based on you experience, how do data analysts differ from data scientists?
I generally lump them in the same category. This is because a lot of companies are saying that they are hiring data scientists, but theyre actually looking for entry-level data analysts. It is a sexier name; people want to become data scientists. I would classify a data analyst as someone who can use business intelligence to generate reports and metrics. Data scientists are capable of doing BI/reporting, but they can use statistic and machine learning techniques to better understand their data and business problems. Ive also found that while a typical data analyst is comfortable gathering data from a pre-packaged source such as a data warehouse, data scientists are much more adept at collecting their own data via web scraping, surveys, etc.
Who do you foresee using your product: analysts, or data scientists? How will it improve the current process?
*Yhat empowers users to deploy production-ready statistical models to the cloud Right now were catering to the data scientists. However were looking to expand and offer a tool that is more accessible to data analysts and business people. Once you go out in the industry, often times youll see data scientists works getting left on the shelf. There are engineering barriers where less technical people cant utilize advanced analytical work. So were building a tool that will enable a company to utilize the prototypes that the data scientists built, and turn them into a deployed data product that can then be used by everyone in the company.
I think youre implying that there is a technical gap between the new tools and the current generation of analysts. To ll this gap, will the tools become easier to use or will analysts have to become more technical?
I think itll go both ways. Having a technical analyst is very helpful. Nonetheless tools have also gotten better, particularly with python and R (to a lesser extent). Tools are becoming mainstream and easier to use. Ultimately, the idea of becoming data scientists is something data analysts are going to aspire to, thus pushing them to increase their technical abilities. It is getting easier for analysts to make the jump into technical analysts and eventually data scientists.
31
There is also a wide spectrum of tools being built: from a excel plugin-ins to high-end machine learning platforms (namely GoodData, MapR, Cloudera etc). I strongly believe that there is no such thing as data science in a box. You cant just take data and stick it into an algorithm that will predict everything (think Google prediction API). Therefore the humans in the process add tremendous value; these individuals provide domain expertise and make judgment calls. No matter where the industry goes, I think that both side of the equation (tools and analysts) are important.
So do you think that this space will always require human interaction?
Yes, I think certain activities will get easier and become automated, but were not going to get to a point where plug in data science software happens.
From a recent graduates point of view, how do you develop the skills to work as a Data Analyst?
It depends on the industry and how technical you want to be. Learning R/ Python will never be a bad thing. The other aspect is picking up a project and doing it. That is the best way to pick up new skills. Without a defined goal, it is challenging to put in the extra effort to complete the project. For example, here is what I did when I was learning R: I picked a project, which was to automate our companys quality control process. Up until then, they were manually made in excel. Ultimately pain and suffering lead to learning.
A lot of people have the idea that domain specic knowledge is more important than the statistics itself. Do you agree?
This actually comes up a lot when talking to colleagues: is the data more important or the algorithm more important. I personally think that there is no replacement for domain specific knowledge. I previously worked for OnDeck, where we used technology to improve the way small business loans were made. I learned that even if you have a bunch of data about the individual, it didnt necessarily improve the lending process. What really matters is asking the right questions, and finding the right data to answer them.
32
simply e-mailed me looking for a project to work on and started volunteering his own time. That is a great way to get exposure; startups are going to be open to people looking for work experience. Reach out and dont be afraid. When we were starting Yhat, we emailed 20 companies seeking advice. Almost everyone was willing to meet with us. We got 10-15 meetings on the calendar during the first few days. People are generally very helpful.
33
ALI SYED
F O U N D E R A N D C E O AT P E R S O N T Y L E
A L I is a data science enthusiast and transformation leader with more than a decade of experience assisting organizations in using insights and intelligent predications as a value amplifier. He loves drinking good coffee, reading great books and meeting interesting people. P E R S O N T Y L E was born from the idea of creating a platform for the people, by the people, to share passion, knowledge, theory and practices of scientifically analyzing data. Persontyles fundamental belief is that helping people learn and apply Data Science enables freedom, and yields important development benefits for individuals and society. Persontyle offers a comprehensive portfolio of Data Science learning opportunities and services for organizations (profit, non-profit, and government) to deliver breakthrough impact and meaningful value using data.
34
As someone who is offering coursework in data science, what do you think is the most important skill that an analyst has to acquire?
Fundamentally, what sets a creative data scientist apart from the rest is an obsession to challenge the status quo. Great data scientists never compromise with mediocrity. Not just data scientists, but all of us should be able to use data to ask relevant questions and challenge the way things have always been done. If we dont break the rules then the rules will break us. There are too many mediocre and boring things in life to deal with and data science shouldnt be one of them. Data Science starts with creative imagination. Our ability to imagine and think beyond the obvious is one of our extraordinary powers as humans. It is why we are different from other beings on the planet. We are creatures of action; we build and create things. We dont live in the world directly; we live our lives through our imagination, which is full of ideas, concepts, theories and ideologies. One can learn tools and algorithms to deal with this data deluge and get hired based on skill-sets reflected on a resume, but you will only be able to add real and meaningful value based on your ability to improvise, adapt, and create. Last thing we need is a breed of data scientists who are too skills-centric and have not dwelled much on the idea of being the biggest skeptic of data. Before we end up with an army of tech-savvy data scientists who are nothing more than clever calculators, we need to ensure that data science remains a science and does not degrade into some gibberish practice that is more of a collection of techniques rather than a discipline. To conclude, if data science doesnt require thinking, challenging, questioning, and rebelling, then its just a useless tech to protect status quo and mediocrity.
Why did you decide to approach educating students in data science via in-person classes rather than an eLearning course or MOOC?
I was expecting this question. In the last 6 months Ive been asked this question so many times that at one point I was seriously thinking about putting it on the back of my business card. I have a short answer and a long answer. Short answer is more positive and long answer includes detailed criticism.
35
We encourage people to attend MOOCs and use any possible channel (i.e. do whatever it takes) to get as much information as you can. However I do think that the classroom style learning will always remain the best format for educating minds. MOOCs are not there to challenge the existing educational channels (universities and the like), but rather to complement them. Just like how eBooks will never replace books, MOOCs will never replace formal educations. And the day this happens, we may bid adieu to future and humanity. There are more than 2 billion potential learners around the world today, and more than 70% of them are unable to afford college degrees. So we have to try all the possible means and channels to make education affordable and accessible to those people. MOOCs and eLearning are one of the ways to address this challenge. Also MOOCs are exciting people into trying new disciplines that they might have not previously had access to.
What is your opinion on using higher education to gain the skills necessary for Data Science?
Schools sole purpose is not to train us for industry. The thought that the sole purpose of education is to prepare us for industry is incorrect, limited and very capitalistic. Going to school is an absolute must; we wont ever encourage a young professional not to go to school. Having said that, the challenge for all of us is making sure that the education we deliver is not commoditized.
Since your company is based in London, have you found that the market for data scientists to be different in UK than in Silicon Valley?
Data science is an emerging discipline and has the potential to become a profound branch of knowledge. In the last year or two, we witnessed universities create departments and offer programs related to data science. This is not just happening in the US or Europe; in fact its happening everywhere. In Asia, Turkey, Middle East, just to the name the few. Data, as a fabric of this digital age, underpins all that we do. Using data as a strategic resource is essential for all, regardless of where you are geographically. So its definitely not just restricted to the Silicon Valley.
How do you think the role of a data scientist will change in the next 5 years?
I cant say anything about future with certainty. I can only speculate. I think
36
37
D AV E G E R S T E R
V P O F D ATA S C I E N C E AT B I G M L
D AV E is the VP of Data Science at BigML. He attended Harvard and also has an MBA from Haas School of Business. He has taken various roles at companies including Yahoo, Cloudera, Groupon and more. B I G M L offers a highly scalable, cloud based machine learning service that is easy to use, seamless to integrate and instantly actionable.
38
What are the challenges for BigML when creating these tools?
The biggest overall challenge is design: were trying to take this complicated modeling process and make it simple for the end user. One specific challenge is making sure users dont overfit their models, so weve paid a lot of attention to making model evaluation really easy. With BigML, you can evaluate a model
39
using a holdout set in just a few mouse clicks. Personally, Id like to reduce that to zero mouse clicksjust automatically run an evaluation in the background every time someone trains a model.
Would you say that BigML reduces the amount of statistical knowledge required to do advanced analyses?
I think there will always be demand for people who understand the theory behind statistics and machine learning. But in the future, these highly qualified people wont spend their time on tasks that are automatable. If its automatable, then why do it manually?
Would you say that data analysts are one of the key audiences for BigML?
Yes, definitely. A data analyst has a good understanding of the business and wants to test out their hypotheses, but they dont know the details of how a decision tree algorithm works, nor should they be expected to. An analyst does need to understand the concepts behind predictive modeling, and I think the simplicity of BigML helps with this understanding. Ultimately, our goal is to liberate the analytical brainpower of these folks who already know the business, the data, and the right questions to ask.
There is a McKinsey report that predicts a 1.5M shortfall in data-savvy analysts and managers. What do you think is contributing to that shortfall?
Part of the challenge is that its hard to learn practical data skills in an academic environment. For example, the skills required to wrangle data into usable form arent something you learn in academia, even if youre pursuing a
40
quantitative degree. Youll undoubtedly learn a lot of valuable theory, but you probably wont learn how to write a complicated SQL query.
41
COMING SOON
with interviews from...
Peter Norvig
Director of Research at Google
Hod Lipson
#7 Most Powerful Data Scientist by Forbes
Hal Varian
Chief Economist at Google
L E A R N I N S I G H T I N D ATA http://www.teamleada.com/