Why is data science sexy? It has something to do with so many new applications and entire new industries come into being from the judicious use of copious amounts of data. Examples include speech recognition, object recognition in computer vision, robots and self-driving cars, bioinformatics, neuroscience, the discovery of exoplanets and an understanding of the origins of the universe, and the assembling of inexpensive but winning baseball teams. In each of these instances, the data scientist is central to the whole enterprise. He/she must combine knowledge of the application area with statistical expertise and implement it all using the latest in computer science ideas.
“What advice would you give to someone starting out in data science?”
1 — Chris Wiggins, Chief Data Scientist at The New York Times and Associate Professor of Applied Mathematics at Columbia
“Creativity and caring. You have to really like something to be willing to think about it hard for a long time. Also, some level of skepticism. So that’s one thing I like about PhD students — five years is enough time for you to have a discovery, and then for you to realize all of the things that you did wrong along the way. It’s great for you intellectually to go back and forth from thinking “cold fusion” to realizing, “Oh, I actually screwed this up entirely,” and thus making a series of mistakes and fixing them. I do think that the process of going through a PhD is useful for giving you that skepticism about what looks like a sure thing, particularly in research. I think that’s useful because, otherwise, you could easily too quickly go down a wrong path — just because your first encounter with the path looked so promising.
And although it’s a boring answer, the truth is you need to actually have technical depth. Data science is not yet a field, so there are no credentials in it yet. It’s very easy to get a Wikipedia-level understanding of, say, machine learning. For actually doing it, though, you really need to know what the right tool is for the right job, and you need to have a good understanding of all the limitations of each tool. There’s no shortcut for that sort of experience. You have to make many mistakes. You have to find yourself shoehorning a classification problem into a clustering problem, or a clustering problem into a hypothesis testing problem.
Once you find yourself trying something out, confident that it’s the right thing, then finally realizing you were totally dead wrong, and experiencing that many times over — that’s really a level of experience that unfortunately there’s not a shortcut for. You just have to do it and keep making mistakes at it, which is another thing I like about people who have been working in the field for several years. It takes a long time to become an expert in something. It takes years of mistakes. This has been true for centuries. There’s a quote from the famous physicist Niels Bohr, who posits that the way you become an expert in a field is to make every mistake possible in that field.”
2 — Caitlin Smallwood, Vice President of Science and Algorithms at Netflix
“I would say to always bite the bullet with regard to understanding the basics of the data first before you do anything else, even though it’s not sexy and not as fun. In other words, put effort into understanding how the data is captured, understand exactly how each data field is defined, and understand when data is missing. If the data is missing, does that mean something in and of itself? Is it missing only in certain situations? These little, teeny nuanced data gotchas will really get you. They really will.
You can use the most sophisticated algorithm under the sun, but it’s the same old junk-in–junk-out thing. You cannot turn a blind eye to the raw data, no matter how excited you are to get to the fun part of the modeling. Dot your i’s, cross your t’s, and check everything you can about the underlying data before you go down the path of developing a model.
Another thing I’ve learned over time is that a mix of algorithms is almost always better than one single algorithm in the context of a system, because different techniques exploit different aspects of the patterns in the data, especially in complex large data sets. So while you can take one particular algorithm and iterate and iterate to make it better, I have almost always seen that a combination of algorithms tends to do better than just one algorithm.”
3 — Yann LeCun, Director of AI Research at Facebook and Professor of Data Science/Computer Science/Neuroscience at NYU
“I always give the same advice, as I get asked this question often. My take on it is that if you’re an undergrad, study a specialty where you can take as many math and physics courses as you can. And it has to be the right courses, unfortunately. What I’m going to say is going to sound paradoxical, but majors in engineering or physics are probably more appropriate than say math, computer science, or economics. Of course, you need to learn to program, so you need to take a large number of classes in computer science to learn the mechanics of how to program. Then, later, do a graduate program in data science. Take undergrad machine learning, AI, or computer vision courses, because you need to get exposed to those techniques. Then, after that, take all the math and physics courses you can take. Especially the continuous applied mathematics courses like optimization, because they prepare you for what’s really challenging.
It depends where you want to go because there are a lot of different jobs in the context of data science or AI. People should really think about what they want to do and then study those subjects. Right now the hot topic is deep learning, and what that means is learning and understanding classic work on neural nets, learning about optimization, learning about linear algebra, and similar topics. This helps you learn the underlying mathematical techniques and general concepts we confront every day.”
4 — Erin Shellman, Data Science Manager at Zymergen, Ex-Data Scientist at Nordstrom Data Lab and AWS S3
“For the person still deciding what to study I would say STEM fields are no-brainers, and in particular the ‘TEM ones. Studying a STEM subject will give you tools to test and understand the world. That’s how I see math, statistics, and machine learning. I’m not super interested in math per se, I’m interested in using math to describe things. These are tool sets after all, so even if you’re not stoked on math or statistics, it’s still super worth it to invest in them and think about how to apply it in the things you’re really passionate about.
For the person who’s trying to transition like I did, I would say, for one, it’s hard. Be aware that it’s difficult to change industries and you are going to have to work hard at it. That’s not unique to data science — that’s life. Not having any connections in the field is tough but you can work on it through meet-ups and coffee dates with generous people. My number-one rule in life is “follow up.” If you talk to somebody who has something you want, follow up.
Postings for data scientists can be pretty intimidating because most of them read like a data science glossary. The truth is that the technology changes so quickly that no one possesses experience of everything liable to be written on a posting. When you look at that, it can be overwhelming, and you might feel like, “This isn’t for me. I don’t have any of these skills and I have nothing to contribute.” I would encourage against that mindset as long as you’re okay with change and learning new things all the time.
Ultimately, what companies want is a person who can rigorously define problems and design paths to a solution. They also want people who are good at learning. I think those are the core skills.”
5 — Daniel Tunkelang, Chief Search Evangelist at Twiggle, Ex-Head of Search Quality at LinkedIn
“To someone coming from math or the physical sciences, I’d suggest investing in learning software skills — especially Hadoop and R, which are the most widely used tools. Someone coming from software engineering should take a class in machine learning and work on a project with real data, lots of which is available for free. As many people have said, the best way to become a data scientist is to do data science. The data is out there and the science isn’t that hard to learn, especially for someone trained in math, science, or engineering.
Read “The Unreasonable Effectiveness of Data” — a classic essay by Google researchers Alon Halevy, Peter Norvig, and Fernando Pereira. The essay is usually summarized as “more data beats better algorithms.” It is worth reading the whole essay, as it gives a survey of recent successes in using web-scale data to improve speech recognition and machine translation. Then for good measure, listen to what Monica Rogati has to say about how better data beats more data. Understand and internalize these two insights, and you’re well on your way to becoming a data scientist.”
6 — John Foreman, Vice President of Product Management and Ex-Chief Data Scientist at MailChimp
“I find it tough to find and hire the right people. It’s actually a really hard thing to do, because when we think about the university system as it is, whether undergrad or grad school, you focus in on only one thing. You specialize. But data scientists are kind of like the new Renaissance folks, because data science is inherently multidisciplinary.
This is what leads to the big joke of how a data scientist is someone who knows more stats than a computer programmer and can program better than a statistician. What is this joke saying? It’s saying that a data scientist is someone who knows a little bit about two things. But I’d say they know about more than just two things. They also have to know to communicate. They also need to know more than just basic statistics; they’ve got to know probability, combinatorics, calculus, etc. Some visualization chops wouldn’t hurt. They also need to know how to push around data, use databases, and maybe even a little OR. There are a lot of things they need to know. And so it becomes really hard to find these people because they have to have touched a lot of disciplines and they have to be able to speak about their experience intelligently. It’s a tall order for any applicant.
It takes a long time to hire somebody, which is why I think people keep talking about how there is not enough talent out there for data science right now. I think that’s true to a degree. I think that some of the degree programs that are starting up are going to help. But even still, coming out of those degree programs, for MailChimp we would look at how you articulate and communicate to us how you’ve used the data science chops across many disciplines that this particular program taught you. That’s something that’s going to weed out so many people. I wish more programs would focus on the communication and collaboration aspect of being a data scientist in the workplace.”
7 — Roger Ehrenberg, Managing Partner of IA Ventures
I think the areas where the biggest opportunities are also have the most challenges. Healthcare data obviously has some of the biggest issues with PII and privacy concerns. Added to that, you’ve also got sclerotic bureaucracies, fossilized infrastructures, and data silos that make it very hard to solve hard problems requiring integration across multiple data sets. It will happen, and I think a lot the technologies we’ve talked about here are directly relevant to making health care better, more affordable, and more distributed. I see this representing a generational opportunity.
Another huge area in its early days is risk management — whether it’s in finance, trading, or insurance. It’s a really hard problem when you’re talking about incorporating new data sets into risk assessment — especially when applying these technologies to an industry like insurance, which, like health care, has lots of privacy issues and data trapped within large bureaucracies. At the same time, these old fossilized companies are just now starting to open up and figure out how to best interact with the startup community in order to leverage new technologies. This is another area that I find incredibly exciting.
The third area I’m passionate about is reshaping manufacturing and making it more efficient. There has been a trend towards manufacturing moving back onshore. A stronger manufacturing sector could be a bridge to recreating a vibrant middle class in the US. I think technology can help hasten this beneficial trend.
8 — Claudia Perlich, Chief Scientist at Dstillery
“I think, ultimately, learning how to do data science is like learning to ski. You have to do it. You can only listen to so many videos and watch it happen. At the end of the day, you have to get on your damn skis and go down that hill. You will crash a few times on the way and that is fine. That is the learning experience you need. I actually much prefer to ask interviewees about things that did not go well rather than what did work, because that tells me what they learned in the process.
Whenever people come to me and ask, “What should I do?” I say, “Yeah, sure, take online courses on machine learning techniques. There is no doubt that this is useful. You clearly have to be able to program, at least somewhat. You do not have to be a Java programmer, but you must get something done somehow. I do not care how.”
Ultimately, whether it is volunteering at DataKind to spend your time at NGOs to help them, or going to the Kaggle website and participating in some of their data mining competitions — just get your hands and feet wet. Especially on Kaggle, read the discussion forums of what other people tell you about the problem, because that is where you learn what people do, what worked for them, and what did not work for them. So anything that gets you actually involved in doing something with data, even if you are not paid being for it, is a great thing.
Remember, you have to ski down that hill. There is no way around it. You cannot learn any other way. So volunteer your time, get your hands dirty in any which way you can think, and if you have a chance to do internships — perfect. Otherwise, there are many opportunities where you can just get started. So just do it.”
9 — Jonathan Lenaghan, Chief Scientist and Senior Vice President of Product Development at PlaceIQ
First and foremost, it is very important to be self-critical: always question your assumptions and be paranoid about your outputs. That is the easy part. In terms of skills that people should have if they really want to succeed in the data science field, it is essential to have good software engineering skills. So even though we may hire people who come in with very little programming experience, we work very hard to instill in them very quickly the importance of engineering, engineering practices, and a lot of good agile programming practices. This is helpful to them and us, as these can all be applied almost one-to-one to data science right now.
If you look at dev ops right now, they have things such as continuous integration, continuous build, automated testing, and test harnesses — all of which map very well from the dev ops world to the data ops (a phrase I stole from Red Monk) world very easily. I think this is a very powerful notion. It is important to have testing frameworks for all of your data, so that if you make a code change, you can go back and test all of your data. Having an engineering mindset is essential to moving with high velocity in the data science world. Reading Code Complete and The Pragmatic Programmer is going to get you much further than reading machine learning books — although you do, of course, have to read the machine learning books, too.”
10 — Anna Smith, Senior Data Engineer at Spotify, Ex-Analytics Engineer at Rent the Runway
“If someone is just starting out in data science, the most important thing to understand is that it’s okay to ask people questions. I also think humility is very important. You’ve got to make sure that you’re not tied up in what you’re doing. You can always make changes and start over. Being able to scrap code, I think, is really hard when you’re starting out, but the most important thing is to just do something.
Even if you don’t have a job in data science, you can still explore data sets in your downtime and can come up with questions to ask the data. In my personal time, I’ve played around with Reddit data. I asked myself, “What can I explore about Reddit with the tools that I have or don’t have?” This is great because once you’ve started, you can see how other people have approached the same problem. Just use your gut and start reading other people’s articles and be like, “I can use this technique in my approach.” Start out very slowly and move slowly. I tried reading a lot when I started, but I think that’s not as helpful until you’ve actually played around with code and with data to understand how it actually works, how it moves. When people present it in books, it’s all nice and pretty. In real life, it’s really not.
I think trying a lot of different things is also very important. I don’t think I’d ever thought that I would be here. I also have no idea where I’ll be in five years. But maybe that’s how I learn, by doing a bit of everything across many different disciplines to try to understand what fits me best.”
11 — Andre Karpistsenko, Data Science Lead at Taxify, Co-Founder and Research Lead at PlanetOS
“Though somewhat generic advice, I believe you should trust yourself and follow your passion. I think it’s easy to get distracted by the news in the media and the expectations presented by the media and choose a direction that you didn’t want to go. So when it comes to data science, you should look at it as a starting point for your career. Having this background will be beneficial in anything you do. Having an ability to create software and the ability to work with statistics will enable you to make smarter decisions in any field you choose. For example, we can read about how an athlete’s performance is improved through data, like someone becoming the gold medalist in the long jump because they optimized and practiced the angle at which they should jump. This is all led by a data-driven approach to sports.
If I were to go into more specific technical advice, then it depends on the ambitions of the person who is receiving the advice. If the person wants to create new methods and tools, then that advice would be very different. You need to persist and keep going in your direction, and you will succeed. But if your intent is to be diverse and flexible in many situations, then you want to have a big toolbox of different methods.
I think the best advice given to me was given by a Stanford professor whose course I attended a while ago. He recommended having a T-shaped profile of competence but with a small second competence next to the core competence, so that you have an alternative route in life if you need it or want it. In addition to the vertical stem of single-field expertise, he recommended that you have the horizontal bar of backgrounds broad enough so that you can work with many different people in many different situations. So the while you are in a university, building a T shape with another small competence in it is probably the best thing to do.
Maybe the most important thing is to surround yourself with people greater than you are and to learn from them. That’s the best advice. If you’re in a university, that’s the best environment to see how diverse the capabilities of people are. If you manage to work with the best people, then you will succeed at anything.”
12 — Amy Heineike, Vice President of Technology at PrimerAI, Ex-Director of Mathematics at Quid
“I think perhaps they would need to start by looking at themselves and figuring out what it is they really care about. What is it they want to do? Right now, data science is a bit of a hot topic, and so I think there are a lot of people who think that if they can have the “data science” label, then magic, happiness, and money will come to them. So I really suggest figuring out what bits of data science you actually care about. That is the first question you should ask yourself. And then you want to figure out how to get good at that. You also want to start thinking about what kinds of jobs are out there that really play to what you are interested in.
One strategy is to go really deep into one part of what you need to know. We have people on our team who have done PhDs in natural language processing or who got PhDs in physics, where they’ve used a lot of different analytical methods. So you can go really deep into an area and then find people for whom that kind of problem is important or similar problems that you can use the same kind of thinking to solve. So that’s one approach.
Another approach is to just try stuff out. There are a lot of data sets out there. If you’re in one job and you’re trying to change jobs, try to think whether there’s data you could use in your current role that you could go and get and crunch in interesting ways. Find an excuse to get to try something out and see if that’s really what you want to do. Or just from home there’s open data you can pull. Just poke around and see what you can find and then start playing with that. I think that’s a great way to start. There are a lot of different roles that are going under the name “data science” right now, and there are also a lot of roles that are probably what you would think of data science but don’t have a label yet because people aren’t necessarily using it. Think about what it is that you really want.”
13 — Victor Hu, Head of Data Science at QBE Insurance, Ex-Chief Data Scientist at Next Big Sound
“First is that you definitely have to tell a story. At the end of the day, what you are doing is really digging into the fundamentals of how a system or an organization or an industry works. But for it be useful and understandable to people, you have to tell a story.
Being able to write about what you do and being able to speak about your work is very critical. Also worth understanding is that you should maybe worry less about what algorithm you are using. More data or better data beats a better algorithm, so if you can set up a way for you to analyze and get a lot of good, clean, useful data — great!”
14 — Kira Radinsky, Chief Scientist and Director of Data Science at eBay, Ex-CTO and Co-Founder of SalesPredict
“Find a problem you’re excited about. For me, every time I started something new, it’s really boring to just study without a having a problem I’m trying to solve. Start reading material and as soon as you can, start working with it and your problem. You’ll start to see problems as you go. This will lead you to other learning resources, whether they are books, papers, or people. So spend time with the problem and people, and you’ll be fine.
Understand the basics really deeply. Understand some basic data structures and computer science. Understand the basis of the tools you use and understand the math behind them, not just how to use them. Understand the inputs and the outputs and what is actually going on inside, because otherwise you won’t know when to apply it. Also, it depends on the problem you’re tackling. There are many different tools for so many different problems. You’ve got to know what each tool can do and you’ve got to know the problem that you’re doing really well to know which tools and techniques to apply.”
15 — Eric Jonas, Postdoc at UC Berkeley EECS, Ex-Chief Predictive Scientist at Salesforce
“They should understand probability theory forwards and backwards. I’m at the point now where everything else I learn, I then map back into probability theory. It’s great because it provides this amazing, deep, rich basis set along which I can project everything else out there. There’s a book by E. T. Jaynes called Probability Theory: The Logic of Science, and it’s our bible. We really buy it in some sense. The reason I like the probabilistic generative approach is you have these two orthogonal axes — the modeling axis and the inference axis. Which basically translates into how do I express my problem and how do I compute the probability of my hypothesis given the data? The nice thing I like from this Bayesian perspective is that you can engineer along each of these axes independently. Of course, they’re not perfectly independent, but they can be close enough to independent that you can treat them that way.
When I look at things like deep learning or any kind of LASSO-based linear regression systems, which is so much of what counts as machine learning these days, they’re engineering along either one axis or the other. They’ve kind of collapsed that down. Using these LASSO-based techniques as an engineer, it becomes very hard for me to think about: “If I change this parameter slightly, what does that really mean?” Linear regression as a model has a very clear linear additive Gaussian model baked into it. Well, what if I want things to look different? Suddenly all of these regularized least squares things fall apart. The inference technology just doesn’t even accept that as a thing you’d want to do.”
16 — Jake Porwar, Founder and Executive Director of DataKind
“I think a strong statistical background is a prerequisite, because you need to know what you’re doing, and understand the guts of the model you build. Additionally, my statistics program also taught a lot about ethics, which is something that we think a lot about at DataKind. You always want to think about how your work is going to be applied. You can give anybody an algorithm. You can give someone a model for using stop-and-frisk data, where the police are going to make arrests, but why and to what end? It’s really like building any new technology. You’ve got to think about the risks as well as the benefits and really weigh that because you are responsible for what you create.
No matter where you come from, as long as you understand the tools that you’re using to draw conclusions, that is the best thing you can do. We are all scientists now, and I’m not just talking about designing products. We are all drawing conclusions about the world we live in. That’s what statistics is — collecting data to prove a hypothesis or to create a model of the way the world works. If you just trust the results of that model blindly, that’s dangerous because that’s your interpretation of the world, and as flawed as it is, your understanding is how flawed the result is going to be.
In short, learn statistics and be thoughtful.”
Data is being generated exponentially and those who can understand that data and extract value from it are needed now more than ever. The hard-earned lessons and joy about data and models from these thoughtful practitioners would be tremendously useful if you aspire to join the next generation of data scientists.