Kirill: 00:00:00
This is episode number 401 with Data Science Instructor, Michael Galarnyk.
Kirill: 00:00:12
Welcome to the SuperDataScience podcast. My name is Kirill Eremenko, Data Science Coach and Lifestyle Entrepreneur. And each week we bring you 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.
Kirill: 00:00:44
Welcome back to the SuperDataScience podcast, everybody. Super pumped to have you back here on the show. Today’s episode is all about jobs, how to get a job in data science. So if you’re just starting out into data science or you are quite confident with the basics of data science and you want to progress further and moreover apply for a job and start a career in data science, this podcast is for you. So Michael Galarnyk is very well positioned himself to talk about this. He has completed a master’s of data science at the University of California San Diego or UCSD. And in addition, he actually has really worked in different fields as a data analyst and data scientist. Plus he teaches data science. He’s taught data science at the Stanford University Continuing Studies. And he’s done online courses. He’s taught through LinkedIn Learning, and many other places. So he’s got the perspective from all different areas. And in this podcast, we talked about quite a few topics.
Kirill: 00:01:54
So we spoke about tools and technologies that you will need to get started into data science, how to stand out, how to build a portfolio. What’s the difference between traditional education, online education and boot camps. Teaching data science and how you can get in on that as well. And finally, soft skills in data science. So quite a few topics covered. And once again, if you are just starting out your journey into data science, this podcast is for you. And on that note, let’s dive straight into it. Without further ado, I bring to you data science instructor, Michael Galarnyk.
Kirill: 00:02:37
Welcome back to the SuperDataScience podcast everybody. Super pumped to have you on the show. And today for this episode, we’ve got Michael Galarnyk joining us from Redmond Washington. Michael, how are you doing and welcome to the show.
Michael: 00:02:52
Good. Thanks for having me. I really appreciate it.
Kirill: 00:02:56
Yeah, man. It’s been a long time. We’ve been trying to have this episode for four months now, it’s crazy.
Michael: 00:03:02
It’s been a while. But I mean, also, just the online education business is very busy right now, as I’m sure you know.
Kirill: 00:03:11
Yeah, yeah, absolutely. And thank you for your patience. I’m actually super pumped about this episode. We’ve got a ton of questions from our community that have come in through LinkedIn. Well, a good amount of questions that will keep us busy for the episode. And we’re talking about jobs. Are you pumped?
Michael: 00:03:31
Oh, yeah. I mean, jobs are really about how you can apply data science to your real life and also support yourself.
Kirill: 00:03:37
Yes.
Michael: 00:03:37
You need to eat.
Kirill: 00:03:40
Absolutely. Absolutely. So for those who are listening carefully, we’re going to be discussing how to get hired in data science. But before we get started, Michael, tell us a bit about yourself. Why are you the person to talk with about getting a job in data science?
Michael: 00:03:57
So I’m Michael Galarnyk. I’m a machine learning with Python instructor for Stanford’s version of adult education. They call it Continuing Studies. I’ve also created a couple of courses for UC San Diego’s Extension School. So again, adult education. And those courses at UC San Diego are machine learning fundamentals, data analytics using Python. I’ve created a couple of courses on LinkedIn learning, data visualization with Python or for Python. And a couple of machine learning ones and data science job courses. So ones like 15 tips to get a data science job which will be available on LinkedIn Learning. One is a remote version of the same course focusing specifically on how to get a remote job.
Michael: 00:04:41
And part of the reason why I give this advice is oftentimes, I aggregate advice from people that are hiring managers or industry leaders or even people that are very famous on social media platforms. So a lot of advice I give is not just my advice necessarily, it’s advice I’ve aggregated from multiple sources.
Kirill: 00:05:05
Absolutely. That’s the way to go. And man, that is incredible. From what I understand, you’re still so young. And you’ve already created so many different courses for so many different universities and platforms, in addition to having a very interesting background. Tell us a bit about your background. I can see from LinkedIn you even worked at NASA as a mechanical engineer. That’s crazy.
Michael: 00:05:29
Well, a mechanical engineering intern. And it got extended from there. So I have a really weird background. My undergraduate degree is in nano engineering. So-
Kirill: 00:05:41
Nano engineering.
Michael: 00:05:41
Yeah. Very, very, very small things.
Kirill: 00:05:44
Nice.
Michael: 00:05:44
And in the course of my undergrad education, I realized I liked the programming aspect, what I was doing a lot more than the actual experimental. So a lot of what I worked on was micro motors. So things that are 10 negative sixth. So things that are like the width of your human hair essentially. Motors repelling around different liquids. Doing red blood cells via ultrasound. Just very, very niche projects. And in my undergrad, I was lucky enough to publish much papers within fantastic research group. And then for my masters, I decided I want to do something that was probably more programming related and something that was more applicable to the job market.
Michael: 00:06:36
So I’ve been to the point where I realized I didn’t have enough skills for the jobs that I wanted. So I went back to school and got a master’s in data science from UC San Diego. And then during that time, I also blogged about data science, and a bunch of different things.
Kirill: 00:06:52
Wow. Wow, really cool. So yeah, and now you’re on your path to… What are you focused on right now? You just focus predominantly on teaching data science. Is that correct?
Michael: 00:07:06
Yes, right now. Before I also, when I was finishing my masters also two years afterwards, I worked at Scripps Research Translational Institute and a lot of what they work on, I mean, they work in a lot of different things. But a lot of what my work was on was machine learning for wearables. And just working with wearable data. So if anyone’s ever worn a Fitbit that has steps, sleep, heart rate, I worked on some studies with my old boss, Giorgio Quer. I always butcher his last name. A lot of it was analyzing long-term trends in sleep activity and heart rate. And they’re actually doing some work with COVID these days trying to predict early outbreaks. So very interesting work.
Kirill: 00:07:54
Wow. Wow, very cool. That’s a lot of converging technologies, wearables, nano devices. How far are we away on data science, of course? How far away are we from nano robots traveling in our bloodstream and curing us?
Michael: 00:08:14
I think it might not be necessarily nano robots, but things that are manufactured for the nano scale that are actually going to be what people treat. So there’s a field called nano medicine, and all that is about using the outside of red blood cells as coding for medicine. So a lot of traditional, like chemotherapy for example, it’s not exactly targeting just the cancer. It ends up killing a lot of your healthy cells as well. So a lot of nano medicine is about specifically targeting cancer, but also the cancer vasculature, or the places where cancer gets nutrients as well. And I think that’s where it’s going to be more likely to go. I think the-
Kirill: 00:09:00
[crosstalk 00:09:00].
Michael: 00:09:00
… actual nano robots that you see here, I think for quite a ways away from those ones specifically.
Kirill: 00:09:06
Okay. Got you, got you. All right. Awesome. Are you subscribed to the Data Science Insider? Personally, I love the Data Science Insider. It is something that we created, so I’m biased, but I do get a lot of value out of it. Data Science Insider, if you don’t know is a free, absolutely free newsletter, which we send out into your inbox every Friday. Very easy to subscribe to, go to www.superdatascience.com/dsi. And what do we put together there? Well, our team goes through the most important updates over the past week or maybe several weeks, and finds the news related to data science and artificial intelligence.
Kirill: 00:09:46
You can get swamped with all the news, even if you filter it down to just AI and data science. And that’s why our team does this work for you. Our team goes through all this news and finds the top five. Simply five articles that you will find interesting for your personal and professional growth. They are then summarized, put into one email and at a click of a button, you can access them, look through the summaries. You don’t even have to go and read the whole article. You can just read the summary and be up to speed with what’s going on in the world, and if you’re interested in what exactly is happening in detail, then you can click the link and read the original article itself.
Kirill: 00:10:24
I do that almost every week myself, I go through the articles and sometimes I find something interesting I dig into it. So if you’d like to get the updates of the week in your inbox, subscribe to the Data Science Insider absolutely free at www.superdatascience.com/dsi. That’s www.superdatascience.com/dsi. And now let’s get back to this amazing episode.
Kirill: 00:10:45
Okay. So the way I want to structure this podcast is to jump straight into the questions from our audience and then each question will open up like a new topic for discussion and we can dive deeper into it. How does that sound to you?
Michael: 00:11:04
That’s fantastic.
Kirill: 00:11:07
Awesome. Okay. All right. So we’ll post the link to the questions on LinkedIn if anybody wants to see them. That will be in the show notes. But here we go. So a question from Dibjot. I hope I’m pronouncing your name correctly. So the question is, I wish to ask what upcoming technologies for a successful… Basically, what are the upcoming technologies a successful data scientist needs to learn? And also, the second question came from Ravi, who said, my question is, number one, what are the latest machine learning or data science tools technologies to learn to be competitive in the job market? Good place to start. Michael, what are the best, the top technologies and tools for data scientists to learn to be competitive?
Michael: 00:11:58
I think it’s not necessarily even the latest stuff. It’s learn R or Python. Choose one and learn it really well and go from there. I think this is very common advice that you hear people like Hadley Wickham and Emily Robinson also say that you just need to learn one tool and go from there. If you try to learn everything, you might be stretched very thin, not be great at any one thing when you first start out.
Kirill: 00:12:26
Okay.
Michael: 00:12:27
Yeah. So that’s my basic advice for tools.
Kirill: 00:12:31
Okay. Well, speaking of R or Python, let’s get that one out of the way straight up. R or Python. Which one, if I can only learn one, which one would it be?
Michael: 00:12:42
Definitely Python. I teach a lot of Python courses, but hear me on this one. So a lot of R is typically, at least traditionally very academic. So a lot of that is by statisticians, by people that typically have a master’s or a PhD. You’re more likely to get jobs and ask for higher educational requirements for jobs typically use predominantly R. So my advice is learn Python because you can do a lot wider range of tasks with Python, whereas R is, while it’s expanding, there’s less usage of R in the industry.
Kirill: 00:13:23
Okay. Why is R traditionally more academic?
Michael: 00:13:28
I would love to know the answer for that one. I think Python has been used for years and years and years before it became such a heavy language for machine learning and data science. So people use Python for web development, people use Python for just a wide range of applications, whereas R has been typically a statistician’s language. So why? It could be how R was developed. It could be just traditional usage, and what libraries were built off of R. But like anything these advices that we give on, learn R versus Python, these things change over time. If someone’s has, it builds a great library, these things can change pretty quickly.
Kirill: 00:14:14
Yeah, yeah, that’s true. But at this stage, I would agree. I’ve been noticing how Python is becoming more and more the go-to data science tool. So I’ll start there. But more broadly, so somebody who’s going into data science and which area of data science would you say, or areas are important? Because you can’t grasp everything right away. And even R versus Python, well, how about visualization tools, Tableau, Power BI, Qlik Sense and so on. How about data storage tools like SQL, or maybe even big data, tools, Hadoop, things like that. So should somebody go straight for Python on the machine learning thing, or should they look around and consider other pathways in data science, such as maybe visualization or data preprocessing and other or data storage even? Other aspects. What are your thoughts on that?
Michael: 00:15:17
So if you were to ask me six months ago, my answer would be quite a bit different than this now. So my traditional advice has always been look for jobs in your area, and see what the job listings are asking for. So look at the jobs and say, “Oh, is this job asking for Power BI or Tableau as well as Python or something else?” And then based on what the jobs are asking for, learn the tools. And these days with the market as it is, being so remote, which is a great thing in a lot of ways, it gets a little more murky. A lot of jobs still want you to be in-person when this ends. So you can also just look at jobs that would be in your area and see what you should learn, and then go backward from there. And as far as visualization tools, it’s great to know, to be able to dashboard in Power BI or Tableau. But the simple answer I know, it’s a cop out, is again, just look at job listings and see what you should learn.
Kirill: 00:16:22
Okay. Interesting. And so then once a person, let’s say they identify, “Okay, I want to do the,” or, “the company I want to work for or I’m interested in,” hopefully both, “in the machine learning aspects of data science. So I want to learn Python, as per Michael’s advice.” And then there’s still in there, okay, there’s the traditional algorithms like, I don’t know, k-means clustering, KNN, Naive Bayes, there’s deep learning, there’s ANNs, CNNs, RNNs. There’s more advanced stuff like reinforcement learning, artificial intelligence and different cutting edge and not cutting edge algorithms there. There’s also natural language processing and things like that. So where does one start?
Michael: 00:17:17
So this is, again, probably a very chalk answer, I always recommend the Coursera Machine Learning with Andrew Ng, hopefully I said his name correctly. There’s a lot of reasons for this. A, it’s a very well known course. So all the material there is stuff that not only is good to learn, but it’s also stuff that people get their interview questions from. So it’s very much a machine learning 1 on 1 kind of class. And then from there, you can learn what you’re more interested in, and then go from there. And then additionally, once you take that sort of class, you can go to other classes.
Michael: 00:17:54
So one class I really like is on Udemy, it’s Python for Statistical Analysis. Because I find that if you want to go the statistical route and get a good grasp on P values, which are very important for a lot of tech companies with AB testing, that’s a great class. And I think actually, that’s one of your classes.
Kirill: 00:18:12
I was thinking that. I think it’s by Sam Hinton. He created it, and we helped him popularize the course. So he’s an astrophysicist. He’s been on the podcast twice now. Yeah, that class I love it. I just out of curiosity, watched one lecture and I couldn’t stop, I watched like three or four in a row. You know the guy was on Survivor. Do you know Survivor?
Michael: 00:18:37
Oh.
Kirill: 00:18:39
The TV show, when they only go on an island and they like… And he was on the Australian version of Survivor and he was like, he cracked up some of the corniest jokes there. Everybody loved him for his very different astrophysicist personality. And it was just so funny. And then he brings that to the course. I love the guy, he’s really fun. How do you know about the course?
Michael: 00:19:06
Oh, I saw the class. I heard the class need me because as an instructor, we typically look for other materials to not only learn from it, but to see where we can get better as instructors. So I found that course because a couple of students mine recommended it.
Kirill: 00:19:23
That’s awesome. I’ll tell him about this. That’s so cool.
Michael: 00:19:26
As far as other classes, as far as getting started, once you take the Coursera Machine Learning, and by the way, for the Coursera Machine Learning course, I don’t necessarily recommend doing the assignments because they’re still an Octave or Matlab, which almost nobody uses in the industry these days, relatively speaking at least. And then from there, if you want to learn Python, there’s the Coursera Python for Everybody Specialization. If you want to learn about practical knowledge of machine learning algorithms there’s a Udemy class called Deep Learning A to Z. If you want to learn [crosstalk 00:20:00]-
Kirill: 00:19:59
That’s our course.
Michael: 00:19:59
… aspect and then Machine Learning A to Z is really good as well.
Kirill: 00:20:06
No way you’ve taken all of those.
Michael: 00:20:08
Oh, I’ve taken parts of each class. I haven’t gone through any class all the way through these days, just based on the busy schedule.
Kirill: 00:20:20
And that’s the way to do it. When you go to a blog, you don’t go and read all the blog posts from start to finish that they’ve ever had. And moreover, that’s the way to read self help books. You don’t read the whole thing, you just go and pick and choose what you need.
Michael: 00:20:33
Yeah, exactly. And as far as books for people starting out, it’s the same thing. You don’t read the entire Pattern Recognition and Machine Learning book by Christopher Bishop. That’s the very good machine learning theory with quite a bit in math. You don’t just read that as like a novel you read front to back so to speak. It’s the same thing with courses.
Kirill: 00:20:59
That’s true. So that’s a lot of cool resources to get started. What about for someone who’s already quite confident in the basics of data science and machine learning, what are the tools that will help them really stand out? Tools or techniques that in this day and age, in 2020 that people should be focusing on?
Michael: 00:21:27
Well, a lot of people want to get into deep learning and that’s a good field to get into. There’s quite a bit of a learning curve, although that might be changing. So I always recommend-
Kirill: 00:21:39
Why might that be changing?
Michael: 00:21:41
So as you’ve seen and you’ve read, so there’s the Deep Learning for Coders with fastai. And a lot of them like fastai applications, fastai is a wrapper for PyTorch among other things. And a lot of the people these days, are trying to make deep learning more accessible for people. So trying to make it so you can do, in the case of Jeremy Howard’s and Sylvain Gugger’s book, I’m not very good names. I’m butchering everyone’s name on the podcast, not great.
Kirill: 00:22:12
That’s all right.
Michael: 00:22:14
The Deep Learning for Coders with Fastai and PyTorch. They specifically designed it so one doesn’t have to have a PhD to understand things. And back in the day, if you used early versions of TensorFlow, for example, you have to specify everything and build everything from scratch. And those were very hard to do. So these days, it’s made it a lot easier to work with Keras and TensorFlow, for example. And Keras and TensorFlow didn’t used to be one project either. So for those that are listening, Keras is essentially a wrapper for TensorFlow in a lot of ways. Keras can wrap a lot of different things. But Keras makes things a lot easier to work with for deep learning than TensorFlow on its own.
Kirill: 00:23:06
Okay. Okay. Got you. But I was talking more about should somebody go into the space of computer vision or natural language processing or reinforcement learning? What’s the hottest thing right now?
Michael: 00:23:20
I’d recommend less on what’s the hottest thing right now but what people’s interests are in and what their background is. So if you have a background in linguistics, for example, you might want to look into natural language processing, because that might be what aligns with your background and interests. Because no matter what you do for a day job, you should probably want to enjoy it or have some interest in a lot of what you’ve done before. And computer vision again, depends on what you like working with. So if you’re interested in recognition tasks, and those sort of things then go into computer vision. Because each of these fields are very, very hot. And I know NLP seems like it’s the hottest thing right now, based on all the transformer models, but again, if it’s what you’re interested in, go for it, if not, I recommend different applications of machine learning.
Kirill: 00:24:20
Got you. Got you. And I like that idea of finding at least one thing that’s hot now that you’re interested in and just learning it and doing maybe some projects in it and putting that on your CV or on your LinkedIn or whatever else, even if the company you’re applying for isn’t interested in that or even if you’re not going to be doing that for your job, it just attracts attention. It shows the hiring managers or recruiters that you are interested in being on top of the latest technologies and then whenever you are in a employed situation and a role, then you will do the same for whatever you’re working on.
Michael: 00:25:09
Absolutely. That’s probably the best advice that people can have for looking for jobs. Even just understanding machine learning and deep learning better. Or even just any visualization or Python task. It’s always the answer to the question, how do you get experience if you need experience to get a job? And if there’s an answer to that the answer is projects. And projects are probably the best substitutes for work experience. So if you don’t have work experience and the job you’re looking at needs three years of experience or five or whatever, in specific fields. If you have a project that can help lessen the gap of experience that you need for a job. If you’re looking for a job, work on projects, not only are the projects good to put on your resume, but they also in some ways are simulations of case studies. Which you often get for job tasks. So if you apply for a job and they send you a dataset and, here do something in this dataset. If you worked on a project before, you can use some of that learning and transfer it over to a case study, for example.
Kirill: 00:26:18
That’s great advice. And let’s move on to that topic. So building a portfolio, we’ve got a question from Kristopher, who asks, what is the best way to build a portfolio and show you can do the job?
Michael: 00:26:33
So one of the best ways to build a portfolio is first of all, you have to find dataset, or even create your own dataset. So the biggest part of building a portfolio and typically with data science, to do any sort of data science, machine learning, deep learning, you have to have some dataset, at least in theory for most projects. So the first thing is to find a dataset, look for something interesting about the dataset. So choose some inherently interesting dataset. So try to avoid the very typical datasets, like Titanic, Iris, MNIST dataset. And sometimes you can look on Kaggle, you can look online, you can scrape data, you can use an API like Twitter. Twitter’s not the most interesting data necessarily, because a lot of people use it. But there’s tons of APIs available and tons of things you can do with data. So get the dataset, and then try to find something interesting to look for in that dataset.
Michael: 00:27:33
And I think the second part of the question is how to showcase your projects. And there’s a lot different ways. So I mean, in a lot of ways you post things on GitHub, you can read a blog based on your results for your project. And a lot of data science is about communicating results to others, and sharing things to other people. So if you’ve read a blog based on your analysis, from a Kaggle kernel there’s a lot of things you can do to showcase your work.
Michael: 00:28:06
One other thing you can do, and this is something that is somewhat controversial, you can also share your work online on Twitter. You can post something on Reddit if you really want to, like a link to your work and ask for feedback. Because if you post something online, there’s a good chance that someone that’s a hiring manager will look at your work. People online aren’t always the nicest necessarily, but there’s a good chance you’ll find something that’s genuinely helpful, and will you the same questions or same questions that a hiring manager would normally see or ask when they look at your project. And in my experience it’s been very beneficial.
Kirill: 00:28:49
Yeah, give us an example.
Michael: 00:28:51
So when I was applying to Scripps Research, back in the day, I was still in school and I used to post a lot of things I would do online and part of my work at Scripps Research was in some small way, communicate results, visualizations and a lot of the work I did for my blog really ended up helping me in a lot of ways because it showed employers that I could actually do what I said I could do. It’s one thing to talk about doing something, another to show or to have public proof of it.
Michael: 00:29:25
So I had a dashboard, online, I think it was San Diego Hearts. I think it might be still up. Back in the day sandiegohearts.github.io. It was for a hackathon and they wanted to see that I could communicate results, visualize data. Then it had some interest in public health because a lot of what I worked on was public health. So I really think it ended up helping me just because it showed I can communicate results from a dataset, showed I was interested in the topic and that’s also what I looked for when I had interns at Scripps Research. It was really nice to see that they did some programming outside their time in school or work or whatever.
Kirill: 00:30:16
Well, that’s really cool. That’s really cool. I often see people say, “Oh, it’s so hard to get a job in data science. How do you get a job in a field that’s so competitive? There’s so many people applying,” and so on. But really, it’s one of the hottest professions on Earth right now. And one of the most in-demand. Of course there’s going to be a lot of supply, of course there’s going to be a lot of people who want jobs.
Kirill: 00:30:44
But if you do what you just said now, if you take some projects, do the analysis, write out the report, present your findings all in a nicely curated, well structured Medium blog or LinkedIn blog or on a GitHub page or whatever. You actually go through an effort and you spend a few weeks doing that per project and you do five of those. And you use that as your go-to, that should be at the top of your CV or at the top of your LinkedIn or you just send that to recruiters or hiring managers. If people do that, you get a job in no time. Seriously, just real fast, you will be taken. You’ll have so many applications coming in, your head will start to spin.
Kirill: 00:31:33
I’ve seen that happen. Even in a field that’s not even as hot as data science, this was in, I think it was around 2013. I was at Deloitte, I was in the data analytics division but I went to consulting for some projects, the consulting branch of the world. And they told me, then showed me this guy applied for a consulting job really competitive. It was, I don’t know, 50 applicants per position at Deloitte or something really competitive to get in there. But they showed me this one guy, who really wanted to jump in Deloitte Consulting. And I think it was digital consulting. So something to do with web projects consulting and things like that.
Kirill: 00:32:20
So what he did was, he set up a website, where when you go to that website, say, his first name surname.com, where when you go that website, you land on a page that is designed fully with JavaScript, that is a game. So it’s like how Google has these Google Doodle games sometimes, if you click on them? So he designed something like that. But it takes up the whole page, and it’s him jumping through different challenges, and it’s like how I’m going to basically depicts his journey of getting a job at Deloitte and he’s going to add value and how he’s going to slay all these monsters and bring a lot of profit. So he made a game about that. And it’s very relevant because the job was about digital and websites and stuff like that.
Kirill: 00:33:06
So all he had to do was, he didn’t have to send a resume, he didn’t have to send his LinkedIn, he didn’t have to apply for the formal process. All he had to do is send that link of a short message to the hiring partner or the hiring director or manager and it’s easy to find their emails, you just take the first name dot, the last name, in many companies. It’s easy to figure out what the email would be. You just send it over and says, “Hey, whatever,” what I would imagine I would say is, “I’m sorry to get in touch with you directly. I saw this position, I think I would be perfect, just have a look at this website I’ve built with for it.”
Kirill: 00:33:43
And when you go in there and you see this whole thing a person’s put so much effort, it’s tailored to this business, to their needs. It’s talking about them, it’s talking to the hiring manager directly. There’s no question about it. Why would you spend thousands of dollars, countless number of hours going through interview processes with other candidates and seeing the best one, when you have one right here. You just have to do one or two interviews, boom, he’s got the job. So imagine doing that in data science. There’s so many companies that would love to get a person that dedicated.
Michael: 00:34:18
Yeah, it shows that you can do the job before you actually have the job. And that’s absolutely a beautiful example of how to really impress people and get a job. It’s not to say that people don’t get a job without building portfolio. It’s just, and then as the market gets more and more competitive, it helps to really find a way to showcase yourself and stand out. Especially now that there’s been quite a few layoffs with COVID, more than quite a few, for a lot of people it’s very hard to get a job. So in times when there’s less jobs, it really helps to do more to showcase your work and showcase yourself as an applicant.
Kirill: 00:34:58
Absolutely.
Michael: 00:34:59
One really quick thing, I just want to mention as far as reaching out to people like hiring managers, a lot of people try to connect and network on LinkedIn. And there’s different ways to ask someone for a referral. And a referral is basically an employee vouching for you and recommending you for a job. And one thing I see a lot is people ask in a LinkedIn message like, “I’m looking for a job. And I’d like to know if there’s any openings in your department.” Just to someone random at a company that you’re interested in. It’s not so much an optimal approach. It could work. But people also have to consider how the request will be taken by the other person that you asked of. Well, a generic request, you want someone to stop what they’re doing and respond to you. And in my experience-
Kirill: 00:35:51
[crosstalk 00:35:51].
Michael: 00:35:51
… it typically works a little bit better to first search for job opening at that company, or the person you’re asking and see what the job title is. Because if you’re just asking like, “Oh, I’m looking for a data scientist job,” and at the company, it’s called applied scientist, it shows you’re not really doing as much research before you’re asking for a job. Because it makes the person that you’re asking a favor for, to do some work to be like, “Oh, are you interested in applied scientists?” OR, “This data analyst role?” Rather than making the person do work, I recommend doing a little work on your end. And also say why you’re interested in the role and why you’re a good fit for the job especially.
Kirill: 00:36:34
Yeah, yeah. Don’t waste other people’s time. Because people have such good BS meters, they’re going to see right through it. You’ve got to be serious about what you’re applying for.
Michael: 00:36:46
Yeah. And do some research on the company. And this is one thing I didn’t used to do. I didn’t used to look up what people typically ask for interviews. So if you go for an interview, for example, a look on Glassdoor, see what people have been asked in the past in that company. Also see, research the company what metrics are they trying to improve if it’s a data science role. What are they trying to showcase? What they typically want in a dashboard if you do visualization. Basically put some research in before you’re looking for a job. And that’s a pretty big piece of advice that I wish I knew early on in my career.
Kirill: 00:37:25
Yeah, that’s good advice. Let’s move on to an interesting question about education, about learning data science. So Francisco, or maybe Francisco asks, my questions are what advantages do you think a classical education in a physical university has for someone learning data science? And you would be the right person to speak to that because you yourself have a master’s degree in data science. And the second part of the question is, what advantages do you think online education has for someone learning data science? And again, it would be great to speak to that because you teach data science online and you’ve learned online as well. So what are your thoughts on that?
Michael: 00:38:08
So one thing I really like about… I’ll get to every part of this question, by the way. One thing I really like about online education is it’s more modular. You can pick what specific subjects are interested in and learn more about those subjects and less about the subjects you’re not. And the reason why this is really important, as even if you want to learn everything about data science, oftentimes when you’re working on a project, and you need some advice on a project, or you need to learn more about specific subject, like for your Machine Learning A to Z course, if I need to learn more about support vector machines, for example, because I don’t know much about them, I can do some research on them. And then when I want to learn about something else, I can pick and choose parts of your course and go from there. That’s something I really, really enjoy.
Michael: 00:38:56
The problem with online education on its own, and this is something that’s being changed quite a bit, is a certificate on its own will rarely get you a job. Again, it’s all about showcasing yourself building a portfolio. And the other issue is it’s hard to be self motivated if you just do online courses for quite a few people. So if you’re more motivated by things that you have to do, like deadlines and those sort of things, there’s some advantage in doing a master’s or a bootcamp, for example.
Michael: 00:39:32
A master’s degree is great. However, sometimes the price tag can be quite high, and it can take quite a few years to get. And when there’s an economic boom, being in school for a couple of years, is more difficult because you’re sacrificing money in a lot of ways. You can do a part time master’s degree or a part time undergrad degrees, take classes, that sort of thing. And one of the downsides of graduate degrees is they’re not always typically job focused. They’re very much on learning theory. And while that’s great, I mean, oftentimes you need to get a job and you need money now. It’s waiting for things… It’s hard. It’s very, very difficult. Additionally if you’re thinking about doing a PhD, it’s just a lot, a lot of time to dedicate to a PhD. And a PhD can be five years, it can be eight years, it could be three in some cases, depending on where you are. So it’s a very long period of time where you’re not making the kind of income that you may need or want in life.
Michael: 00:40:42
I will say one thing about masters and PhDs that I really do like and even undergrad degrees, is how do you get experience if you need a job to get experience. It goes back to that question. And one really wonderful thing about being in master’s, undergrad or PhD program is you can intern while you’re in the program. And that’s one way to build up experience while not necessarily having a full time job. And whereas when you’re in a bootcamp, or if you’re doing an online certificate program or just taking a class online, a lot of employers are not necessarily open to someone learning a subject and interning. Because the barrier to entry for interns is a lot lower. And at least in the US, most of the time for a data science internship, you’re getting paid. So it’s a way to make money when you’re going through.
Michael: 00:41:31
One thing I really like about bootcamps and this has been mentioned by a lot of different people, is there’s benefits and drawbacks. One of the biggest benefits of a bootcamp is that bootcamps have a vested interest in getting you a job and-
Kirill: 00:41:44
Why is that?
Michael: 00:41:46
Well, it’s good for their reporting numbers. If people do well in bootcamp getting a job, then that makes that bootcamp look better. So that means that they typically help you with the resume, portfolio building and they make some effort to connect you with your employers. So if they place people at Google, for example, or Facebook or any company really, it makes them look better, and people are more likely to recommend the bootcamp to friends or their employers.
Michael: 00:42:12
Bootcamps can be a little bit expensive sometimes and time consuming, especially if it’s a full time program. And with anything, I’m just talking about right now, I would recommend people sit down, research the quality of individual courses online, graduate degrees and bootcamps. Because it’s not just a master’s degree or bootcamp, it’s which master’s degree and which boot camp? So look at alumni that have gone through your program, look at your instructors to see if they’ve been successful in something that you’re interested in. And that’s really, really important.
Kirill: 00:42:51
That’s very cool. So you’ve got a couple of options actually, not just online education and university but you also have bootcamps and that’s like a faster version compared to universities, right?
Michael: 00:43:04
Absolutely. And bootcamps, it used to be back in the day you needed a PhD for a data science role, or at least a lot of people would ask for one. And these days, those requirements are being knocked down in a lot of ways. They ask for less of that and just more of, can you do the job, do you have the skills? And a bootcamp, typically, a good bootcamp at least, provides the skills to do a job.
Kirill: 00:43:31
Yeah, absolutely. Absolutely. That’s very cool. That’s very cool. So there we go. That’s the answer to that question. And another interesting one, so I really believe in teaching as a way of learning something even better. And I’m very excited to see that Fahad asked the question, which is the following, if someone would like to teach data science to a younger generation like school students, how can someone prepare the materials? Or in which way can he approach them? He or she. Well, in which way can pretty much anyone? I guess he’s asking for himself, but anyone approach them?
Michael: 00:44:15
So one of the biggest things is the age group. So this advice could be completely invalid for older generations of people. People that are in their 20s now versus 16 versus 12 versus 8. So for the younger crowd, you might want to first lay off the Python and learn Scratch. Or I fare well with people typically when I’m doing that, like programming Madecraft, for example, just get people into programming, necessarily. Not necessarily data science and machine learning when they’re really, really young.
Michael: 00:44:48
And then also, and this goes with an older audience too, tell people about the application of what they’ll be learning, other than just in theory. So rather than these are multiplication tables. Tell people what they can do with multiplication, for example. Same thing goes for data science. So why are we learning about linear algebra? How does this relate to the world at large? If we’re teaching linear algebra to college students, for example. Teach them about how eigenvectors and eigenvalues are used in real life, if you’re interested in dimensionality reduction, later on. That’s actually a terrible subject to talk about. But talk about how concepts relate to real things. And that’s probably the most important part, is the application and the intuitive knowledge of how something works. Not necessarily, like, here’s the hard and fast math, because you lose a lot of students that way.
Kirill: 00:45:47
Very interesting. So how do you go about preparing materials for your course?
Michael: 00:45:52
Very time consuming, as I’m sure you know, for your own courses.
Kirill: 00:45:56
Absolutely. Absolutely.
Michael: 00:45:59
It goes also with blogs. There was a Medium blog, ages ago written by people from Medium about, which blogs are the most popular and most well looked at. And for blogs, for example, a lot of it’s just how much effort you put into them. How many hours do you dedicate to a blog, for example. It’s the same thing for a course for example. The courses that tend to do better are ones that people put more time into.
Michael: 00:46:25
So for me, oftentimes for a course it’s material that I either struggled on or a material that makes students ask me for. And that’s more common these days. If a student ask me, “Hey, I’m not understanding box plots,” for example, and let’s say 20 students have the same issue, then I find out what they’re having issues with, and then work my way backward and write a material on that or write a script for a video, and that sort of thing.
Michael: 00:46:57
I should also note that if you’re interested in teaching data science, oftentimes, writing tutorials is actually a good way to build portfolio as a data science instructor, or a Python instructor or a R instructor. The way that I got in touch with this podcast, for example, is my friend John David, got me in touch with you. And the way John David knows me is because of people from Madecraft, who I work with for LinkedIn Learning, they saw one of my blogs online, asked me to make a LinkedIn Learning class. That’s how I met John David and that’s how I met you.
Kirill: 00:47:33
So, wow. That’s a cool story.
Michael: 00:47:35
Yeah. So it really has helped to write tutorials or make YouTube videos or whatever. It is not perfect. I have terrible YouTube videos online, for when I was first starting and they actually did help me in a lot of ways, become better at teaching. So also, if you’re interested in teaching, I advise start making content and you’ll learn from your experiences and make your content even better. So I think the biggest thing is to just start doing something and then figure out what you’re doing good and bad along the way.
Kirill: 00:48:09
For many people the hurdle to just starting doing something is the imposter syndrome and fear of not knowing enough or being an expert, and we’ve spoken a lot about that on the podcast before with other guests. What I want to talk to you is, for many other people, the roadblock for just doing it is this feeling of it can’t be bothered. Or it’s so much effort. It’s just, I’ll do it tomorrow, that kind of thing. How did you overcome that?
Michael: 00:48:56
Well, at least for me, it was first about accepting that I won’t know everything. And there’s always a subject I don’t know. And as an instructor that’s hard to tell your students, “Oh, I just don’t know that. I don’t know that advanced topic or I can’t derive a Kernel PCA for you on the spot.” And this is about stuff that you don’t know everything. And for students that are just starting, I always recommend you don’t have to dive all the way in something when you first start. You can do 20 minutes a day or an hour a day. And pick one course to learn from or one project work on.
Michael: 00:49:33
And there’s a lot of being overwhelmed because learning everything in data science would be very, very overwhelming. So pick one specific thing, learn Python, learn R, focus on learning one model for machine learning or like one BI tool if you want learn data visualization and you want to use a BI tool, start just learning Power BI or started just learning Tableau. And then once you feel okay with that, or you don’t want to do it anymore, learn something else.
Kirill: 00:50:08
Yeah. One step at a time, right?
Michael: 00:50:10
Yeah, always one step at a time. The biggest advice that you can give people, it’s just start somewhere and-
Kirill: 00:50:18
Otherwise it looks so big, right?
Michael: 00:50:19
Yeah, exactly.
Kirill: 00:50:20
You probably have, I have this feeling, tell me if you have the same. If I looked back, I don’t know, the journey I had in the past five years, and I was to know this whole journey five years ago, I would have been so scared, I might have not just gone ahead with it. But when you break it down into little steps, it’s much more appealing and it just looks doable.
Michael: 00:50:46
Exactly. There’s no way I could have done what I’ve done if I just saw everything I was like, “Oh, I’m going to do all that. That seems impossible.” And also you don’t always know where your choice is going to take you. I don’t think you thought you’d have a massively popular podcast, or that your classes would be so popular on a lot of different platforms.
Kirill: 00:51:15
Yeah, yeah. Yeah, absolutely. Absolutely. I mean, it’s just baby steps for sure. One more question we’ve got that I really want to cover off before we wrap up, and that is from Francisco, again, who says, what do you think are the necessary soft skills to be a successful data scientist?
Michael: 00:51:37
So you’re asking somebody who doesn’t have perfect soft skills here. So I think one of the biggest things is communication. And that’s a cop out answer again. But learning to listen to people, find out what they want. And also find out how to best communicate results to people. And a lot of data scientists listen to people, listen to stakeholders find out what the problem is, identify that and then try to find a way to a solution. And I think one thing that I’ve learned more of is to listen to people about what exactly they’re asking for and also to ask new clarifying questions on the narrow down exactly what they’re looking for. So communication is a big one.
Michael: 00:52:29
Also just don’t be too difficult to work with. And the more skilled you are, you can be probably not as great with the soft skills. But a lot of who you, when you interview often most people are looking for, can I work with this person? Is this person going to be painful to work with? So likability and just not necessarily be easygoing, but just listen and learn to communicate better of your results.
Kirill: 00:53:05
That’s good. Empathy. Level up your empathy.
Michael: 00:53:10
So I have a question for you. So-
Kirill: 00:53:12
Sure.
Michael: 00:53:12
… what are the best soft skills in your life as a data scientist?
Kirill: 00:53:18
Gosh, best soft skill. I would say two. One, learning or understanding how to ask the right questions. A lot of the time people don’t really know what they want. So when somebody asks you for a data science project, it’s a whole art on its own, to go and clarify what exactly is it. Not just what they want, what is it that they need. So you need to deliver to people not just wants, but also uncover their needs as well. And there’s a lot of communication that is explicit, there’s also a lot of implicit of communication or a lot of things that on the onset of the project that might be assumptions they’re making that you need to dive in and understand, “Hey, that’s not really a fact, that’s an assumption.” So at the onset of the data science project, absolutely important to do that. Because otherwise you might be doing a whole wrong project. In fact I’ll correct that, not two, but three things. So that’s the first one.
Kirill: 00:54:23
The second one is throughout the project another important soft skill is to constantly communicate your progress, and check up with the stakeholders to make sure they’re happy with your progress. Because if you just leave it until the end, let’s say it’s a six week project, and then on week five, you show them, “Hey, this is my result I’m going to present to you in a week,” and they’re like, “Well, hold on, that’s not what we wanted.” Boom, you’re in trouble. You have one week to fix a five week project. Whereas if you communicate with them from the start, you go and sit with them, you watch them do their job or you walk around with them on the factory floor or whatever it is you’re working on, and you constantly update them on the results, you turn them into an advocate.
Kirill: 00:55:14
And that way they are going to, now at the end of the six weeks, they don’t have a choice but to be on your side, when together you and them are presenting to their boss, to the upper management, to the final client or so on. So create yourself an ally along the way and also avoid any wrong turns in the project.
Kirill: 00:55:36
And then the final third part of, what is it called? Soft skills in data science, perhaps, they’re all important, but this one is the one that will really turn you into a rock star, in-demand data scientist that everybody’s after. Headhunters are calling, recruiters are emailing. And this one is presentation. The art of standing in front of an audience, whether it’s 12 executives or 200 people in a big hall and having your slides up there, you can do the whole project right, you can do all the steps right, but if your presentation style and skill is not engaging, is not fun, your delivery will be dry, and people are not going to be hooked.
Kirill: 00:56:27
The thing is, humans are not just robots. And even if you give a human the best possible advice if it’s just pure logic, in most cases, they won’t take it. There’s tons of books on psychology written about this. Humans make decisions based on emotions, not on logic. They later justify the decision to themselves with logic. But they make on emotions. How many times has anyone listening to this been to a store, you get super hyped about something, you buy it and then you can home and you’re like, “Why did I buy that?” Because you can’t justify it with logic. There is no logic behind you, you bought it based on emotion. And so same thing with presentations, you need to present. You did all this work. And data science is like pure logic, pure data, pure mathematics and analytics and logic in a way. So you did all of that. But now you have to switch your brain, you have to completely shift into the emotional domain, you have to present it from an emotional standpoint, you have to convince the decision makers to act on your insights from an emotional standpoint and then if they ever want that justification or confidence, you have the data behind that.
Kirill: 00:57:43
But you should put that aside and you should turn it into a fun engaging presentation. And that’s like acting. There’s a whole acting part that you have to augment your data science career with. And only if you have that, then will you be one of the top 1% of the data scientists in the world that really not only can deliver the results, but do the math or calculations and data analysis but can actually move the needle, make their company go forward. And that’s ultimately how you’re going to be measured, or how people are going to see that you’re making a difference not by the analytics that you do, but by the decisions and actions that are driven by your analytics. So that would be my take on soft skills.
Michael: 00:58:36
That is a fantastic answer. And I’m glad you gave that. Because yeah, you have to be able to-
Kirill: 00:58:41
Thanks.
Michael: 00:58:43
… you do all the hard work of the analyzing, the long nights, the early mornings, in some cases of doing all the manipulation, some case machine learning, visualization. And if you can’t present your results, it might be the best model in the world but if it doesn’t provide value, or people don’t understand the value, then it’s almost useless in a lot of cases.
Kirill: 00:59:07
Yeah, absolutely. I also agree. Well, Michael, this is a good part for us to wrap up on. I would like to thank you for coming on this show and sharing your insights. And I’m sure a lot of people are going to find this useful, especially those starting out into the space of data science. Before I let you go, please, could you tell us where people can find you, follow you or get in touch?
Michael: 00:59:35
Well, you can find me on Twitter @GalarnykMichael. You can also connect with me on LinkedIn, send a message request. You can find my blog on Medium. And I think it’s also probably Galarnyk Michael. You can take any of my LinkedIn Learning classes. Any way you want to contact me, feel free to. And I’m looking forward to hear from people, hopefully.
Kirill: 00:59:57
Awesome. Awesome. So we’ll definitely all those links in the show notes. And Michael, one more question, what’s a book that you can recommend to our listeners? Some book that’s changed your life?
Michael: 01:00:10
Well, I’ll start with the machine learning aspect of it, Python Machine Learning by Sebastian Raschka. Hopefully I said the name correctly. I really like the book just in terms of just content in general. I think it’s just a really well thought out book and the author put a lot of time into it. And as far as non data sciency books, I really liked The Name of the Wind by Patrick Rothfuss. When he wrote it back in the day, I’m not sure if he’s actually coming out with a third novel. It’s a fiction book. But I really sometimes like listening to audiobooks when I’m either working out or on a plane ride or driving or whatever. But yeah, those are my books. And there’s many more [crosstalk 01:00:54].
Kirill: 01:00:55
What’s the second one called, again?
Michael: 01:00:57
The Name of the Wind.
Kirill: 01:00:59
The Name of the Wind. Interesting. What kind of fiction is it?
Michael: 01:01:04
A young boy goes to school, family gets killed, [inaudible 01:01:11] the best of it.
Kirill: 01:01:13
So modern times not futuristic.
Michael: 01:01:15
No, no, no. Fantasy. Heroic fantasy.
Kirill: 01:01:20
Okay, got you, got you.
Michael: 01:01:21
Yeah. I mean, the problem with the book and this is a double edged sword, is the third book’s not out. The second book in the series was released in 2011 and it’s 2020 now.
Kirill: 01:01:33
Yeah.
Michael: 01:01:34
So with any series, I usually recommend waiting till it’s done. Because you never know if the author is not interested anymore in the book or busy or whatever.
Kirill: 01:01:44
Got you, got you. Awesome. Well, Michael, amazing. Thank you so much, again for coming to the show. It’s been a pleasure.
Michael: 01:01:52
Okay. Thanks for your time.
Kirill: 01:01:59
So there you have it everybody. Hope you enjoyed this podcast and got some valuable takeaways from here and got some clarity on what the next steps are for your career. Some of the things that we spoke about might sound like common sense, like very straightforward. But when put together and executed on properly, this is how you get the top jobs. Top paying, most exciting, most open jobs for career growth, for learning the best jobs in the market, you need to execute these steps in combination.
Kirill: 01:02:36
My personal favorite part from this podcast was when Michael talked about the difference between traditional education versus online education. He said there that online education has this property of being modular. So you can pick out what you want. And in the traditional education, you actually can’t. You have to go through the semesters or years in a sequential order and just yes, you can maybe sometimes pick subjects, but often, they’re already selected for you. And you just have to do them, even if it might not be extremely relevant to you. So I had never thought of it that way before. And that shows another advantage of online education.
Kirill: 01:03:17
And there were some questions that we necessarily didn’t get to, or I didn’t read out specifically in the questions that were submitted on LinkedIn, because we covered them off in the conversation. But some people were asking, how do you get a job without a formal education? Well, you don’t really need formal education. You can get away with online education or get away if you can actually excel with online in this education space.
Kirill: 01:03:43
And also liked what Michael mentioned about bootcamps as another alternative. So if your online education is just not enough for you, or you’re finding it hard to be motivated, then bootcamp, you don’t really have to go for a full degree at a traditional university. You can if that’s what you want, but also remember there’s another option called bootcamps, which are perhaps shorter than the full degrees, which sometimes can take years. But at the same time, they’re more involved and will force you to continue and motivate you a bit more than just online courses.
Kirill: 01:04:23
So there we go. That’s the breakdown. And as always, you can get the show notes for this episode at www.superdatascience.com/401. That’s www.superdatascience.com/401. Woo hoo, we’re passed the 400 mark. And there you’ll find the transcript for the episodes, any materials we mentioned plus URLs to Michael’s LinkedIn, Twitter, and so on. So make sure to connect and follow Michael see what he gets up to.
Kirill: 01:04:51
And one final thing, if you know anybody who’s just starting out into data science and you think they need some clarity, they might be a bit lost in all the things that are going on in this space, send them this episode. It’s really easy to share, just send them the link www.superdatascience.com/401. And thanks again for being here. I’ll see you next time. Until then, happy analyzing.
