Database Reference
In-Depth Information
Gutierrez:
What do you look for in other people's work?
Tunkelang:
I love to find people with good taste in problems. For me, worthy
problems are more interesting than clever solutions. As data scientists, we are
truly in a position to change the world, as we can improve people's health and
well-being, optimize allocation of resources, guide better policy decisions, and
similarly worthy problems.
I'm inspired by people who work on inspiring problems. Jeff Hammerbacher
once said, “The best minds of my generation are thinking about how to make
people click ads. That sucks.” I wholeheartedly agree, and so that's why
I suggest that people should focus their best talent on worthy problems.
Gutierrez:
What does it take to do great data science work?
Tunkelang:
Hilary Mason and Chris Wiggins said it best: A data scientist is
someone who obtains, scrubs, explores, models, and interprets data.
1
Which
means, as Drew Conway expressed in his Data Science Venn Diagram, that
data scientists need to be armed with hacking skills, math and stats knowledge,
and domain knowledge.
2
And, perhaps most importantly, data scientists need
to have strong critical-thinking skills and a healthy dose of skepticism.
Gutierrez:
When hiring and training people for your group, how do you
approach teaching or mentoring people to develop these skills?
Tunkelang:
Failure is a great teacher. One of my best learning experiences
in college was implementing an algorithm from a paper, only to have it not
perform as claimed. I contacted the authors, who told me how they'd tuned
their systems for each example in the paper. After overcoming my initial
reaction of indignation—after all I'd worked for months on my own compet-
ing approach—I realized that I'd learned an important lesson to not believe
everything I read in a peer-reviewed publication.
As W. Edwards Deming allegedly said, “In God we trust. All others bring data.”
In science, the default assumption is the null hypothesis, which puts the burden of
evidence on the hypothesis you're trying to prove. These are all variations on the
same theme—if it's too good to be true, then don't believe it until you can back
up your belief with data. It's easy and important to tell people to be skeptical, but
I doubt it's enough to overcome our cognitive biases. This is a case where experi-
ence is not only the best teacher, but also perhaps the only teacher.
1
Hilary Mason and Chris Wiggins, “A Taxonomy of Data Science” (September 25, 2010:
2
Drew Conway, “The Data Science Venn Diagram” (September 30, 2010:
