Kirill:
This is episode number 105 and this time we have not one, not two, but three guests.
Welcome to the SuperDataScience podcast. My name is Kirill Eremenko, Data Science coach and lifestyle entrepreneur. Each week we bring inspiring people and ideas to help you build your successful career in data science. Thanks for being here today and now let’s make the complex simple.
[Background music plays]
Hello ladies and gentlemen, and welcome back to the SuperDataScience podcast, I’m super excited because the
DataScienceGo (DSGO) 2017 conference has just completed, I’m literally still in the hotel in San Diego where it all took
place in this past weekend and it was a blast. We had almost 200 people attend and hundreds more watching on the livestream online, we had some amazing speakers, amazing guests. We’ll definitely talk more about that in the coming weeks, but for now what I wanted to do today is share one of the sessions that we had and that session was a panel discussion. On Sunday, day two of the conference, we had a panel discussion where all of the attendees were able to submit questions through Twitter, through a special hashtag and we were answering those questions. Obviously, I was running the panel discussion, I was asking our three guests the questions from the audience, and they were giving their views. We’ve got almost an hour, or about an hour of questions and answers and I think you will find them very insightful, we had quite a lot of laughs and quite a lot of insightful answers. Our three guests at the panel discussion were Ben Taylor who is …
If you’ve been listening to our podcasts you will know all three of our guests. But if you don’t know Ben Taylor, then he is a thought leader in the space of data science, he is a founder of his own company with some partners, called ZIFF.ai where they create artificial intelligence and deep learning products. Very, very motivational person and I follow Ben Taylor myself all the time. I’ve learnt a lot from him.
Also, we had Hadelin De Ponteves who is my very good friend and business partner, he was also in the panel discussion, and he is the co-author along with me of courses such as Machine Learning, Deep Learning A-Z, Computer Vision A-Z, AI, Artificial Intelligence A-Z.
Also in the panel, we had Urie Suhr who has also been on the podcast. Urie is the Head of Talent Acquisition for a company called Collective[i] which is a data science and artificial intelligence consulting company on the East Coast of the U.S. So, there we go, that’s our panel for today and I can’t wait for you guys to learn all these amazing insights from them and have some laughs along with us. Without further ado, I bring to you the DSGO 2017 panel.
[Background music plays]
Kirill:
Welcome to the panel discussion. We’ve got a lot of questions, like a ton of questions coming through, got them here on my phone. Urie, you know everybody. We’ve got Urie, Ben, Hadelin. I think it’s a good mix of skills, expertise, if you have more questions, send them through along the way. Okay, off we go. You guys excited?
Panel:
Yes, so excited, of course.
Kirill:
All right. First question. How do you respond to people who think that AI is in the hype cycle? Ben, you’re the best to start with this one.
Ben:
That’s fine, they can say that but people are making a lot of money. So, money drives everything and there’s success stories. I’ve seen companies who have 10Xed their market capability because of AI. As those success stories come through, that’s not hype, that’s real evaluations, real revenue. Self-driving cars, that’s not hype, like once you start seeing trucks, and deaths going down. It depends on the application.
Kirill:
Thank you. Hadelin?
Hadelin:
Yeah, I think there’s a lot of communication around AI and that plays with the hype and all that. Some of the communication is wrong like I hear saying that AI is exponential. Well that depends on how you see it. Research is not exponential, we still have a lot to progress with research so definitely this will be a long term and for me it will last for another very long time.
Kirill:
Okay. Next question, I’ll probably start with you, Urie, with this one. How can you address work-life balance? How do you transition from working each day to working from home?
Urie:
That’s a fantastic question. Work-life balance, I feel those who are I should start of by saying, are successful working from home. What working from home means is what you guys or candidates feel, or employees I should say, feel most comfortable with the environment you’d be most productive. Interesting fact I know I might come off as an extrovert but I’m actually a mixture, I’m an ambivert, so when I’m in work mode, I love to just hide, and I like to just be in a corner off somewhere and that’s how I work best. I want no distractions and I just don’t want any interaction so working from home really works for me. I guess, you know … Hope I’ve answered that question. To find that balance I think it’s okay to … Am I answering the question? I feel like I’m missing what the question is.
Kirill:
No, that’s good. How can you address work-life balance and more important like the next stage, how do you transition from working each day at work to working from home?
Urie:
I think it’s to ultimately set up your space to feel most comfortable, as far as working from home. But that transition, the coming from working with 20 people to 300-person company and then all of a sudden being by yourself. I think if the question is more about how do I stay connected, there’s always forms of communication like Slack and so forth or to remember to engage with your team. But working from home I think will definitely be more beneficial to those who have a certain work process and a work flow. It’s just about setting that up and getting that process in place, I’ll just reiterate that.
Kirill:
All right. How about you, Ben. You work from home, right?
Ben:
I work from everywhere. Starbucks. (I’m going to plug Starbucks right now). I work at Starbucks a lot, just all over the place where I am. I love Starbucks. I think one of the things that’s changed in my mind is I kind of see this as an efficiency play. It’s not about your hourly efficiency, it’s about you weekly throughput. My first job, working for Intel Micron, it was about buts in seats. I had a complaint come in against me through HR that Ben’s butt is not in his seat. But I was working from different areas, working on different things so we kind of obsessed about our weekly efficiency and we realized I’m much more effective working in a Starbucks or all by myself with my headphones on, the Bose, noise cancelling, and if I need to go skiing it’s more about the weekly efficiency, it’s not about, hey you didn’t work 8:00- 5:00 today. So work-life balance, that kind of goes into you overall emotional state, wellbeing, your relationships with your loved ones, happiness. Because I can imagine a scenario where I work 100 hours in a week and I don’t get as much done as I was able to do in 40 hours. So work-life balance is important.
Kirill:
How does somebody approach their employer with that conversation. Saying that actually I will get more done for you if I will sit in a Starbucks.
Ben:
That’s hard when you’re working for a more traditional company. Once you kind of break into the space and you build out a reputation you’re top of the game, you can ask for whatever you want.
David, my co-founder, when he used to work for companies, he would tell them, I’m only coming into the office once a week, and they didn’t really know how to handle that. But it was, if you want my attention on your problem, I’m coming into the office once a week. He’s a kind of an introvert. He’s more efficient when he doesn’t come into the office. We had another data scientist at HireVue, if you wanted his attention, he demanded four months of ski vacation. People would laugh, we went through attempting to hire him with that on the line, that he needs four months of ski vacation. And he’s still going to make more than most data scientists make with four months of ski vacation. Once you kind of get to the top of the game, if your employer is not working with you, there are lots of employers that will … And you have the perks of the developers like the Xboxes and anything-goes kind of environments so you have to find the right employer. And there are definitely employers out there that are too black and white, they’re traditional, they’ll see these requests as being very, very, strange and out of control but if you’re offering the value it doesn’t matter.
Kirill:
Yeah, gotcha. I can relate to that. I’ve been doing that from kind of … If you are high level, you can demand whatever you want. If you’re like slowly getting to high level, you slowly can demand more and more. I started this habit, it’s a bad habit, I started this in high school. I was like, I’m not coming to school on Wednesdays. I only did school Monday, Tuesday, Thursday, just because I didn’t feel like it, and still finished. Then with my previous employer also at some point I was like, I can get the job done in four days. I’m not going to come to work on Wednesdays. I’m not even going to work on Wednesdays, I’m going to get the job done. We were agreed on salary range, whatever, but I need that time to work on my own projects. If you are valuable to them, they will meet you. So, sometimes, I think Urie mentioned this, you’ve got to have just the courage to have that conversation, what’s the worst that can
happen?
Hadelin, what are the biggest challenges for working from home?
Hadelin:
Yes. I have a terrible work-life balance because I work all the time. Since we are launching a course every month, I always want to not lose a second. Therefore, I only take a break like a big break, going out on the weekend or going for short holidays between each course, between production of each course. I don’t think that’s good. I think I should sometimes tell myself that it’s okay if I don’t progress today, it’s okay if I will progress tomorrow. If I need something to be done today, I can tell myself, okay, I can do it tomorrow. I should really work on that because working all the time is good but sometimes you need to live your life.
Ben:
Just to reiterate, kind of what you had on a little bit. I think if you can convince your employer to hold you accountable on a week timescale or two week-scale, then the fact that you didn’t come into work today doesn’t… Yeah, but I complete the sprint, so focus on the deliverables and not the butt in the seat, hourly.
Kirill:
There’s a good book about it, it’s called The Fourth Economy and it’s also referenced in The End of Jobs by Taylor Pearson. He talks about that the 9:00-5:00 that originally came from the industrial revolution where the amount of time you spend at work is proportional to your output because you are on a conveyor belt. We don’t live in that time any more, it’s not about the time, it’s about the effort, the creativity, the ingenuity that you put into your work. You can get… Personally, I know for myself, if I work between the hours of 7:00pm and 11:00pm, in those four hours, I can get done the amount of work I usually get done in two days, just because I’m more productive then. It also ties into a lot of those things. Next one, a data science question for a change. How is machine learning going to meet traditional BI? Ben.
Ben:
BI?
Kirill:
Business Intelligence, yeah.
Ben:
I think it will do some of the discovery. I think it’s funny because sometimes as humans we get excited about creativity like, look what I found or look what I did. But as soon as you see a pattern, a lot of times AI can come in and automate that. So, if I’m going through and I notice some correlation or a regression or some relationship between two variables, with programming and Python, you could easily just crank out an entire report that stack ranks whatever vanity metric you care about.
I think some of that will slowly become more and more automated and a lot of the operators of BI software, they have AI on the roadmap, they’re looking at ways to accelerate the actionable insight, but I think it will be a slow transition.
Kirill:
Gotcha. Okay, Hadelin?
Hadelin:
I think machine learning already met with BI. I think there’s going to be some more and more automation, for example let’s take DataRobot. DataRobot, you just give the inputs it gives you some outputs, I think it could automate some of the processes in the BI problems. Definitely it will bring a lot of contribution.
Kirill:
Thank you. What’s the one critical failure you had in your career and how did you overcome it? Let’s go one by one. Urie, what is your critical failure?
Urie:
My critical failure? It’s like an interview question.
Ben:
You’re on the other side. You have to show a strength. What’s your weakness? I work too much.
Urie:
Actually, I was going to go with… At a job interview I decided to switch things up and actually share my transparent vulnerability and when I was asked a question of what my weaknesses were and what my failure was, is that I had this issue where I overpromised and always felt like I was under- delivering. And it’s because when you want to say, yes, yes, yes, and be that yes person, I know it could be really scary to push back and to feel like you don’t want to have your superior feel like you’re not doing enough or contributing enough or saying no is like a scary thing.
I was very honest on an interview stating that next time around… I should say the failures that I’ve had was because I wasn’t very upfront, and I was transparent and set the expectations and communicate that to my superior. And so some projects didn’t work out the way it should have because I was too afraid to admit that, to admit that these expectations are just above and beyond, and to explain or connect with my superior that let’s try to see what’s more manageable as far as expectations, and being upfront.
Kirill:
Thank you. Ben? Except for that skiing accident.
Ben:
Restate the question so I can make sure.
Kirill:
What’s one critical failure you had in your career and how did you overcome it?
Ben:
I think one of the consistent failures I’ve had, maybe not a failure but hindsight is always 20-20 so you can look back and have regrets. Is ripping the band aid off sooner for different technology stacks. When I was in undergrad, we had a professor that told all of us we should reformat our laptops and install Gen 2 which, that’s actually not a good recommendation. But his recommendation to switch to Linux was constantly said and I ignored him until I graduated my undergrad and then I committed to changing, which was painful. By hindsight, I had lots of regrets that what would have happened if I had switched three years earlier.
Switching to Linux was one, starting to write documents in LaTex, ramping on Vim, and more importantly abandoning MATLAB and switching to Python which was very hard. I’ve talked to some people here, where some of you guys are rock stars with your commercial platforms, but what you don’t realize is your greatly limited on your ability to accelerate cycle time, provide value and stay up to date. So that’s a very hard transition. Hindsight I wish I had transitioned much sooner because now I have regrets of not transitioning sooner.
I think right now I’m going through it with Pycharm, so I do all my development in Vim and my cofounder has switched to Pycharm which I see as being powerful and more useful, but you kind of have this safety net and that will probably be another one when I switch to Pycharm I’ll have regrets, of like well I could have done this …. Yeah, opportunity cost, exactly.
Kirill:
Gotcha. Hadelin?
Hadelin:
I think it was the first try, second try and third try of the RNN we implemented for the Deep Learning course. But now I’m happy to tell you that it’s solved. I re-implemented the model, I re-recorded the RNN part recently, so that’s good.
Kirill:
Okay, thank you. A question for Urie. Have you ever taken leap of faith on a candidate who lacked major skill sets you were looking for, and why?
Urie:
Tough question. I have but let’s define major lack of skill sets. If you are not even a fit for the role or the opportunity, I can’t take that chance because there’s no connection. But if candidates in the past have had that true connection to that role and because they were so devoted and passionate about it, they were already half way in. I knew that it didn’t matter if they lack a quarter of the skill set, they’re going to ramp up and they’re going to get it, and they’re going to exceed far more than anyone who is just like, oh yeah, fancy me, I have all these technology stacks.
Kirill:
An interesting question, one I also wanted to ask. Just came through from you guys. Urie, you talked about this balance in supply and demand and even that there is a report by Burning Glass Technologies saying that job posts for data science roles are open for 45 days, whereas the average is 40 days, so that’s five days longer than average. That talked about the insufficient supply. When do you think this balance will even out, how long will this last?
Urie:
I think it will shift sooner than later. Probably give it, to be fair I would guesstimate probably 72 months at most. Hopefully. Yeah, three years.
[Laughter]
The reason I say this is because companies out there are still trying to transfer and make that process over to really understanding what the landscape of data science looks like. So, don’t be fooled, if they’re not … The Googles or the Facebooks or the Netflix, you know, those guys get it. But for the rest of the world, they are catching up on trying to understand what the requirements really are needed for their particular role that they’re trying to fill in data science. It’s going to take that process to move over but it will.
Kirill:
Ben, do you agree? Higher, lower, what’s your estimate?
Ben:
Yeah, I think that’s a good estimate that it will continue but then there will be some contraction period. I think a lot of that contraction will be driven by software, where I don’t need to hire at entry level because DataRobot or some other provider actually does everything and I can have an intern or some junior analyst who does not have our background doing it. There will always be very competitive jobs for the top tier, so I would encourage people just reckless commitment, just dive head first, get entry level but become an expert in deep learning and what’s sate of the art, and you won’t have to worry about positions and job security.
Kirill:
That’s a very interesting estimate. Don’t you guys think that while companies will figure out ways to solve that problem for themselves and these roles will get filled, don’t you think that there are still tons of companies out there who are not even considering data scientists, and when those companies are going to come on line it’s going to increase as well with time so there will be new roles in those spaces? What do you think about that?
Ben:
I think that’s why like the next three to four years, you’re going to see demand increasing. Where more and more companies … It’s interesting because two years ago we were actually worried that we started our start-up too late, like we missed the wave. And we realise, no that’s not true because two years ago executives were not asking for AI. It was not part of the board meeting discussion. Well now it is, we actually have executives that have decided an AI mandate and you’re going to see that more and more. Well, now that will be the expectation.
Kirill:
So you started at the right time?
Ben:
Yeah, I’m pretty happy.
Kirill:
Question for Hadelin. How would you advise students in a maths major to better prepare themselves in getting a job in machine learning?
Hadelin:
First of all, they are in a maths major so that’s a huge plus
to start in machine learning pretty efficiently. My advice is to
read the best books, again. Even the deep ones like highly
theoretical ones because thanks to the maths background
they have they could understand, for example, introduction
to statistical learning, elements of statistical learning which
are the best books, they could understand also the Deep
Learning by Ian Goodfellow and Yoshua Bengio, so that’s
great. And besides, what I highly recommend once they
understand the basics of machine learning is that they can
go to the research papers. Research papers are full of maths
and you actually need a maths background to understand
them because you don’t have all the details. So, read the best
books, best online resources and research papers.
Kirill:
Cool. Would you agree?
Ben:
I failed my first linear algebra class in college because it was
so boring. No real-world applications. And then years later I
found out on my own that I love it, it’s one of my favourite
topics, but it’s only my favourite if it’s taught through
machine vision and image processing. It’s something that’s
actually tangible. Well, some of these maths departments, it’s
terrible.
Kirill:
It almost sounds like if you’re studying maths. Who’s
studying maths in the audience? Okay, we’ve got a couple of
people. If you’re studying maths then along the way you can
just look for those real-life applications via machine learning
and understand what you’re studying, how is it actually
going to be useful to you in the future.
Ben:
Yeah. And there are so many fantastic examples in image
processing, so everything you guys are covering is eye candy.
Kirill:
Okay. Next question, about the future. Ben, AI and deep
learning in the future and the possibilities that come with it.
Ben:
Yeah. Last year, Baidu was the first company to announce
superhuman speech and Microsoft has announced it this
year as well which means that they have speech nets that are
getting better than a 4%-word error rate and that opens up
so many possibilities, not today, but in the next three years
just like we’re addicted to our phones, we will all be addicted
soon to our personal systems. We’ll have virtual personal
systems that follow us on our phone, in our homes, in our
car, no more note taking, no more …
Siri is not state of the art when it comes to voice recognition
but in the future, you can have more natural conversations
with all of your appliances, where they are not these canned
responses that are expected right now, and I think that’s
going to be wonderful and I think every home in the future
will have a GPU in your home to handle all your video
streams, all audio streams and to support this virtual
system. Maybe someday you’ll fall in love with them like that
one movie Her.
[Laughter]
Which is a great movie. Some people might think it’s weird
but I think it’s funny.
Kirill:
Gotcha. Hadelin?
Hadelin:
I totally agree with Ben. As you said in one of our courses, in
the future, everybody will use machine learning, probably
have a GPU in their home as today everybody uses an iPhone.
So I totally agree with that and I don’t think it’s going to come
in a long time, but in a short time.
Ben:
There is a huge wave happening right now. iPhone X has an
AI chip in it and Apple has made a huge investment in
supporting deep learning, NXG Boost model exports to the
iPhone and I think what’s happening, there is a huge lag
where the iPhone developers have no idea what that even
means. We know what it means. I think in the next two years,
you’re going to see a huge ramp in amazing AI apps where I
don’t have to connect to the network. And you’ll hopefully
have a Siri competitor.
Kirill:
Correct me if I’m wrong. Is this understanding correct? Right
now, if I want to do … For the phone to recognize my speech
and put it into text, I have to be connected to the internet.
Ben
Yeah. Your old phone will still run. Apple has Core ML and
you can export Keras, MX Net, Caffe, you can export all those
models to your current phone but it will destroy your battery.
So you’ll be mad because in 30 minutes your battery will be
dead. Maybe not true but with the new iPhone X, they can
just hum along.
Kirill:
So now they’re in-building these functionalities inside the
phones?
Ben:
Yes. So you could do a better speech recognition on that
phone but you may not appreciate your useability.
Kirill:
Gotcha. Okay. What are the three most important areas to
master to be a great data scientist? How about we all
contribute one. Ben? The most important area that you
think.
Ben:
Do you want to be technical or behavioural or?
Kirill:
Let’s go behavioural.
Ben:
Technical.
Kirill:
Who wants technical? Who wants behavioural? Technical it
is.
Ben:
Maybe we do technical and behavioural. So technically,
absolutely deep learning. It’s prime time, its’ not a fad, it’s
not going to go away. Become an expert in all things deep
learning. Behavioural, I will take the one that we’ve all been
saying, passion. Low hanging fruit, have passion.
Kirill:
Good choice. Urie?
Urie:
I would agree with the same thing. Deep learning, definitely
invest in that, that’s not going to go away, in fact that’s our
future. And two, for behavioural, conceptualization.
Conceptualization, don’t be afraid to be a little crazy because
crazy gets you answers, gets you closer to really actually
defining that accurate picture.
Kirill:
What do you mean by conceptualization?
Urie:
My experience of interviewing data science candidates, one of
the things that many companies actually screen for is just
your thought process. Thought process meaning like how are
you … What your thinking, not ability, is like, but how you’re
conceptualizing, how are you making your correlations, how
does this make sense, and how when people say abstract,
really explain your abstract. And maybe I use crazy in the
term of abstract but it’s okay to try your best to explain that
or practice on how you explain your abstract.
Kirill:
Gotcha. Hadelin?
Hadelin:
Technical, definitely AI, to complement deep learning like for
example the A3C model, deep reinforcement learning that
goes with deep learning but there is the reinforcement
learning part. And behavioural, business sense. We need a
business sense to understand how data science can optimize
the business, how it can help contribute, bring added value
and that’s extremely important. I actually learned that in
Google.
Kirill:
Next question is from people who are just looking for their
first job. How do you find the first job for a fresh candidate?
Where would you look for one? Ben?
Ben:
If I was looking for my first job, the way I approach this
problem is I actually don’t care about my first job. What I
care about is the job I’m going to have in three years. So I
actually might take a job that is not preferred but it’s going
to get me exactly where I want to be in three years, so I may
Show Notes: http://www.www.superdatascience.com/105
take a pay cut, work for an anchor brand and then land my
three year job. Or that might help me sift through the
opportunities. So I’d look for a job that gave me … if it wasn’t
going to build up my reputation to land a bigger job, I would
look for a job that gave me a lot of problem variety. Because
some jobs, you’re going to be very good at just structure data,
or just like one thing and that’s not what I would want. I
would want a job that allowed me to kind of hit on all different
types of problems and they had buy-in from management on
AI, and ideally some mentors at the company. Some
companies have some very strong data science teams that
have reputations, so I’d love to get in with something like that
even if I had to take a big pay cut. Because it’s not about the
pay, the goal is to be where you want to be in three years and
the path to get there may not be intuitive, you have to really
think through that.
Kirill:
Gotcha. Urie, from your perspective where do you look for
candidates? Where is the best place for them to go for you to
find them?
Urie:
I think it’s everywhere really. There’s not one … I know that’s
like, that sucks, what do you mean everywhere? But again,
connection is important and just to piggy back off of that.
Your first jobs should really be about, yes, devote your
passion in it but think ahead. Like, this is where I want to
go, so leverage what you can. I would say as far as just listing
out the opportunities or where you can go, LinkedIn, you
have Glass Door, well, local to New York there’s Built In NYC,
there’s many different types of hiring platforms that you can
show your visibility on. But again, the more organic approach
is always sometimes even the faster approach, just really
connecting directly with the opportunity.
So if you are entering a Kaggle competition, check to see who
you can connect with. Or if you’re connecting to someone
over at Stock Overflow, I’ve connected somebody through
that way as well. For your first opportunity, I would definitely
say start-ups are a lot more … I don’t want to say that they
are looser on their skill sets but they are flexible because they
are in the same position and same boat, too. They’re trying
to launch their start-up. Regardless how it is, whether it be
one, two, three people on a team or a 15-person team, it is a
great opportunity for someone who’s first starting off to say,
okay I have the basics but this is where I have to grow and
ramp up. You are gaining that ability and just because
there’s less of a hand-holding or less of a structure, it actually
makes for a fantastic developer or data scientist. Because
they’ve had to overcome so many battles being in a position
where they have to wear many different hats.
Kirill:
Okay. Thank you. What’s the most memorable interview that
you’ve had with a data scientist? Same question for you, Ben,
you’ve hired tons.
Ben:
Not tons but a handful. Memorable interview. Good or bad?
Kirill:
Either.
Ben:
My old employer, HireVue, they have world’s largest interview
repository of … They do interviews, which means they have
outliers. And they have famous outliers. There’s an outlier
where an individual did an entire interview with one word.
Tell me about your previous experience and your
background, Python. Tell me about what you like to do for
fun, Python. The entire interview, of course the answer was
no but we always kind of joked internally that this guy may
have been a genius with Python.
[Laughter]
Which could have been useful and maybe for that role we
didn’t need social skills and stuff and they could have passed
on a rock star and, we should really go like follow up and see
what happened with this guy. But, you know, you see stuff
like that.
One of the first data scientists I hired, he actually failed one
of the interview questions and that was one of the things that
stood out the most. He didn’t fail it but we expected them to
have natural language processing and we wanted them to
have experience, and there was questions that tell us about
your experience in natural language processing. He said,
well, I actually don’t have experience, I haven’t done any
projects with it, but let me tell you what I know about it. And
his response about what he knew about it was better than all
the others that said they had done stuff like sentiment
models and built stuff out, so we hired him. And we’ve been
thrilled with him, he’s been a fantastic hire, and he’s really
developed and grown and become an expert in all things data
science. So sometimes not having experience but just
showing that you have good breadth and you care about the
space. That was to me the most memorable, him not having
experience was his strongest trait because he compared all
of these data science experience, but I wanted to hire this
person.
Kirill:
That really ties in well with what you said, Urie in your talk.
Just say what you know about it, what you’ve researched
about it. Great example there.
Hadelin, you mentioned that business sense is an important
thing. How do I learn business skills like marketing
operations, supply chain, to be able to relate and apply my
data skills to solve a real business problem? How do you
develop that business sense?
Hadelin:
This time I don’t believe that courses or books are the best
way, I think experience is the best way. I would recommend
to work in companies like consulting companies that do some
data science missions to optimize some business, like to
optimize some cost reduction or revenue optimization. You
don’t have to look for a consulting company, you can actually
do that in every company. Every company has, like, a finance
department or a BI department. Well not everyone, but most
of companies have a finance or BI department and in these
departments, you really develop the business sense and you
can use data science to optimize the business, optimize work
on the BI and I think it’s a great way to develop the business
sense.
Kirill:
Okay. Thank you. Would you agree, Ben?
Ben:
Business sense. Most data scientists don’t have a strong
business sense and that’s hard. How do you get people to
appreciate money? I think one of the negative traits I see is
what I call the academic behaviour. Where someone with a
strong business sense they will kill a pet project and they
won’t overcomplicate a problem, they will start simple. I don’t
know how you teach that.
Kirill:
Do you think that every data scientist needs a business sense
or there’s …?
Ben:
No, you need at least one. One data scientist who is a lead or
managing a group, is essentially baby-sitting really smart
people that can do tasks, but they’re deciding that I know
that problem you’ve been obsessing about I’m actually going
to pull a plug on it because you haven’t delivered value in two
weeks and we have to have you working on something else.
So, not every data scientist needs the business sense but
having the business sense puts you in a position to be that
lead data scientist, that chief data scientist, that chief data
officer. The seven figure salaries coming from this Fortune
100 companies, they need a data scientist that has very
strong business sense and very strong communication skills.
And then they want IP because every company obsesses
about IP. But I think business and communication are …
[Interjection]
What your CEO wants to know? That’s a book, that sounds
like a good book for data scientists.
[Interjection]
Okay. I want to make sure I mention the second one, so Lean
Six Sigma Skills. A third book I really like is The Lean Startup
by Eric Ries, where this was very different from the way I
behaved as a programmer. He encourages you to fail first,
and what can I do to pretend like I’ve delivered value. And
that’s something I see as a huge… Someone with a strong
business sense if I asked you to do a project, you might
actually push back at me and say does the customer even
want this project? Let’s go fake a presentation. Let’s actually
fake results, let’s fake some data science results and show it
to a customer advisory board. That data scientist is gold
because they understand the business sense and they just
protected me from a two month commit mistake which has
happened to me. We committed two months of resources, got
all the way to validation and the customer doesn’t want it.
But internally it sounded like a high five data science idea.
Hadelin:
I would just like to add that I agree with you that not
everybody in the team needs a business sense especially the
data scientists. However, if one data scientist hopes to evolve
like hopes to evolve in management positions for example
lead, I think it would need to start working on developing a
business sense otherwise if a data scientist keeps focused on
his expertise, on his very high technical skills, he will stay in
that position, I think, forever.
Kirill:
Some people are happy to do that. But if you want to grow
you need to develop a business sense.
Ben:
Do you mind if I ask a question? Something I’ve been
obsessing about is I want to hire a data scientist that is
unblock- they don’t get blocked on anything. What’s that skill
set?
Kirill:
Unblocked?
Ben:
I have a data scientist working, they’re working on problems,
working, working, and they hit a wall. Hit a wall where they
can’t advance, they can’t do this, this isn’t working, I can’t
get this piece of code around. But there is another data
scientist, it’s almost like they’re a train, they will just, like
blast through any brick wall even on things they haven’t been
exposed to, like hey here is that mix now, I need you to take
this rake file and stream it through your model. Is that
curiosity? Grit? Curiosity, grit?
They pull in outside resources and cut through red tape to
get stuff done, so what traits do you hire, what competencies
or behaviour would you look for, for me to convince you that
I will be your freight train? It doesn’t mean I cannot come talk
to you, but you can focus on what matters and not have to
unblock me.
Kirill:
I define it as, excuse the French, getting shit done. I actually
put that on my resume at some point, like I’m a person who
I don’t care about data science and everything, one of my first
lines is “I get shit done.” If you see that on a resume, hire
them. It might be my resume. Yeah, now you’ll get a bunch
of resumes, great thing. Don’t put shit, just say I get things,
I get stuff done.
[Laughter]
Ben:
Just send a resume with a hand drawing of a train, breaking
through a brick wall, and then like a picture of you. Or just
like your email.
Interjection:
The reason why people get blocked is they get anxious … You
panic, you don’t think you can finish it so you either put it to
the side or you try and pile it on to someone else. And so the
technique we’ve used before to train people in this situation
because you can’t do that all the time, is to deal with anxiety,
learn to problem solve … It happens with study, the anxiety
of not knowing the answer is what bogs you down. A person
who doesn’t have that pressure or that trigger of anxiety
when they’re under pressure is someone that can generally
solve problems and not panic and not get distracted.
Urie:
I want to add to that, though. Those who do have anxiety but
have learned to channel that anxiety and overcome pressure
also is something that’s also important to look out for.
[Interjection]
Ben:
One of the things we joke about internally is rubber ducking.
If someone asks a question, you’re blocked, you have to come
ask the question, we ask, have you asked the rubber duck?
Because usually just by asking the question it kind of comes
together. I feel like there’s value in asking the rubber duck in
the corner your question and thinking about it. It doesn’t
solve all this but it’s just one thing that kind of helps with …
Kirill:
Another think I’d look for is, like Urie mentioned, like
hobbies, extra-curricular activities. Sometimes when people
have a major accomplishment that is super hard and one of
those, like there’s something they had to work towards for
years and years, and years, like skiing, like, they might put
that I skied this mountain. And you know that that is super
insane to do that, and they must have failed many times
before they got there but they did achieve that. Some people
think of it as bragging, like why would I put that, like I won
the culinary class master gold medal iron chef in whatever
country this year, it’s not about bragging it’s showing that
you can get things done even outside of work and I think as
an employer, I would look at that, I would be like, oh, that’s
interesting, if they made that accomplishment that means
they are a person that can persevere and get through all these
road blocks like a train as you say. So, that could be
something to look for as well.
Ben:
One skill set that came out that we did not think was an
important skill set for the data scientist was our CTI realised
that I was better at googling than one of our data science
members. And why are you good at googling? Because you’ve
googled a lot. I would have never thought that that was a skill
set. Like finding the Stack Overflow, or finding this
information, we spend a lot of time googling.
Kirill:
That’s what we do most of the time when we’re creating a
course. Google, google, google.
Ben:
So you’re expert googlers.
Hadelin:
I think so. I have progressed a lot.
Kirill:
Two things you need to put on your resume, get things done
and expert googler. All the jobs are yours.
[Laughter]
Urie will hire you first.
Ben:
There’s a website called Let Me Google That for You. So if
someone asked you a stupid question, you google it and you
send a link and what happens is it pops up and they watch
you google the answer. I’ve sent it to people and they get
really offended.
[Laughter]
I’m teaching you and showing you the answer. You should be
happy.
Kirill:
Next one. Can anyone be a part-time data scientist? For
example, I might be a doctor or an attorney or a teacher or
realtor, or a mum, or an FT. I don’t know what an FT is. Oh,
a mum full-time, okay. What do you think, Urie, can anyone
be a part-time data scientist?
Urie:
Of course, I’m in HR and I do data science. I still work on my
personal growth in data science and absolutely appreciate
that I’m in a lucky circumstance that my data science team
encourages me in fact pressures me to keep up. But it’s about
what you’re passionate about. If you enjoy it and you enjoy
the subject, it’s still learning, it’s still growth. You never know
where that growth leads you.
I took the course not to become a sensational data scientist
overnight. I took it because it meant something to me,
because data matters to me and I needed to answer my own
personal questions. Why do I over-analyse, how can I use my
brain in conceptualization? It just felt good, that’s why I took
the course. Whatever comes of it, I think it’s just if you enjoy
it, that’s what matters.
Kirill:
Ben, what do you think?
Ben:
I think it depends on the role. I think there’s tremendous
value in people understanding at a high level, at the 30,000ft
level, all things data science related because they already
have the business sense, they’re already fully ramped on the
subject matter, and they can drive the innovation and the
decisions. If I’m hiring for a technical data science position,
one of the major concerns I have for someone who isn’t
completely full-time focused on data science is compared to
the pool of applicants, they won’t have the breadth and they
won’t have the depth that the other applicants have. I think
there’s huge value in having both.
Kirill:
Even like when say somebody wants to be in data science,
they just can’t, what do you think of like freelance work, for
instance? Online freelance work?
Ben:
I would completely count freelance as previous work history.
For someone that… HireVue they have a new policy pretty
much across the board, they will not hire someone without
previous work experience. We want another company to pay
the tuition. And that makes people upset because they say,
well, if all companies do that, what does that do for all of us?
Freelance work, Upwork, any consulting completely counts.
So if you could show someone you’re billing out $150 an hour
for consulting, you can ask for that six-figure salary. You’ve
shown from three, four, five, different jobs you’re billing at
this rate, you’re definitely worth the six-figure salary. Even if
you haven’t landed it. I think that should be encouraging
because it doesn’t matter what your background is. You
could be a dropout, English major, literally it does not matter.
You’ve successfully closed consulting.
Kirill:
That’s great. And guys, Upwork.com if you’re looking for a
part-time
or
getting
that
recognition.
Like,
at
SuperDataScience, we hire data scientists who answer your
question in our courses and there’s an AI guy from Upwork,
there’s a machine learning guy, there’s lots of people, there’s
an R data scientist, there’s a Python data scientist, they all
come from Upwork. It’s a huge marketplace, check it out.
Hadelin, what are your thoughts?
Hadelin:
I absolutely think that it’s possible to combine both, I actually
know a lot of people who do that and that’s because it’s so
easy, unless you have an exclusivity line in the contract with
your full-time job, it’s very easy to have a complementary
freelance job in data science. I know a lot of people who do
that, so as you say they reached a six-figure salary this way,
and thanks to Upwork, there’s high demand and you work
whenever you want, you just need to get the work done so
very easy.
Kirill:
Gotcha. I’ve got an interesting question for you guys to wrap
this up. What would you do if not data science? If data
science didn’t exist, what would you be doing? Can’t say
skiing, Ben. Hadelin would you like to start this one?
Hadelin:
Yes. Artist.
Kirill:
Artist. As in what kind of artist?
Hadelin
Because I believe that the future jobs that will remain in the
future will be engineers and artists, and by engineers it’s
mostly AI developers like machine learning developers, all
working around AI and artists. So if not AI, then an artist.
Kirill:
Like, you’d paint, you would create things?
Hadelin:
No, I would do theatre.
Kirill:
Theatre? Good choice. Lucky we have data science.
[Laughter]
No, I mean like his talent would be lost, data science talent
would not be … Urie, what would you do?
Urie:
Wow, so that’s a tough question because I think it depends
on the day. I would say if it wasn’t this I’d probably be a pilot,
I love flying, just being in the air and also my biggest fear is
fear of heights. So I’m very scared of heights and I think it’s
just one of those things if they scare the most, I want to do
it. So it’d be a pilot. And then some days … When I was very
young what I wanted to be when I grow up is to animate for
Disney. Animation, I think that still fascinates me, like 3D,
or now it’s like augmented reality and all those great things.
[Interjection]
Kirill:
Awesome. Ben?
Ben:
This question is easy. I’d have a fruit orchard that I’d tend
and I would be a python breeder. I would have hundreds and
hundreds of pythons. Snakes. I’d be a snake breeder and I
would have a fruit orchid and I would be…
Kirill:
Where did that come from?
Ben:
There’s like this whole like, it’s not a black market because
it’s legal. People trade ball pythons and apparently the
Germans and the U.S., we love it. So they have $200,000
snakes that are sold back and forth and exchanged. I
wouldn’t sell the $200,000 snakes but I would sell the
$1,000. I would just breed pythons. That would make me
very happy.
Kirill: Guys, if any of you are reconsidering a career in data science, there’s your options. Pilots, artists, snake breeders. Thank you very much everybody. So, there you have it. That was the panel for DataScienceGo2017, I hope you enjoyed those questions and answers and maybe you heard some of the questions that you had personally, answered in this panel as well. On that note, you can find the transcript for this episode at www.www.superdatascience.com/105 Make sure to connect with all of our guests if you haven’t yet. You can find their linked in URLs on that same page. And we look forward to seeing you at the next DataScienceGo conference so next time you can ask your own questions and get answers from our future panel. Thanks a lot for tuning in today, I look forward to seeing you next time and until then, happy analysing.
[Background music plays]