J. B. Mailman
Data Scientist, Northwestern Mutual • Graduate • Data Science Bootcamp • Online
Sep 04, 2021
The Metis DS bootcamp was life-transforming, in the best sense. Perhaps no other three-month period in my life was so distinctive. It was tons of hard work, which I always felt glad to do, all day, late into the night, and most of the weekends too. (If there were more than 7 days in the week or more than 24 hours in the day, I would have spent the additional ones on the bootcamp work.) Ok, so I’m already unusually inclined and motivated to learn (I had a PhD already) and I’ve been keenly...
The Metis DS bootcamp was life-transforming, in the best sense. Perhaps no other three-month period in my life was so distinctive. It was tons of hard work, which I always felt glad to do, all day, late into the night, and most of the weekends too. (If there were more than 7 days in the week or more than 24 hours in the day, I would have spent the additional ones on the bootcamp work.) Ok, so I’m already unusually inclined and motivated to learn (I had a PhD already) and I’ve been keenly focused on learning situations for years (taught college for a decade) but I believe my perspective is relevant to anyone contemplating diving into a Data Science bootcamp.
I carefully chose Metis because of its curriculum, standards, and reputation, as well as the fact it has a careers team (to help with the job search). I was not disappointed in the least. First of all, the best thing about bootcamp generally—which probably can’t be replicated any other way—is the variety of events, activities, and tasks that center around and circle around the subject (actually each subtopic). This was ideal for learning. For instance, often in the morning there would be a relatively small-scale one-on-one problem to work on for 45-minutes or so to build up a toy version of an algorithm from scratch (these “pair-programming” exercises nearly every day were _great_—so much so that some of us continued doing it a few times a week even after bootcamp was over); then a couple of hours later the same day (or sometimes the day before) there would be a one- or two- hour lecture on that same Machine Learning approach, and then later a workshop-ish lecture to get your hands dirty using the scaled-up industrial-strength pre-packaged (not from scratch) version of that algorithm. This meant that we learned how the engine under the hood actually worked so we would become skilled enough to tune it.
The lectures themselves were of various styles by various instructors, which again was bonus. (A uniform or cookie-cutter approach, or having only one instructor, would have been disastrous.) Although doing a bootcamp online (via zoom) might seem like a disadvantage, we turned it into an advantage. by intermittently having on-topic side chats, sometimes with apropos hyperlinks and diagrams posted, questions asked and answered (by the instructor, the TA, or other boot-campers) and the occasional witty nonsequitur for levity to keep the mood light—given the intensity of the material! There were many times when it was surprisingly thrilling. One of the reasons for this also is that people are coming into the bootcamp with such varied backgrounds, so the kinds of questions asked were all over the map; and some bootcampers came with their own expertise (in math, stats, comp-sci, probability, etc.) so they could sometimes contribute added nuances to the instructor’s already expert explanations. The mood and tone was always respectful and collegial—which can’t be taken for granted.
The project-based structure of the course was also ideal because we were grasping to master the lecture/workshop material in order to be able to implement it right away in the individual project we were working on for the given two-three week period, with that deadline looming. I got a taste of all the steps, from locating data, to wrangling/cleaning it, modeling, testing, interpreting, and building up a presentation that would be scrutinized by my instructors and peers. The day-to-day flow of the Metis bootcamp meant becoming fairly fluent with professional tools like git and fluent in installing and learning unfamiliar python packages needed to complete your own particular project.
Learning DS and ML is complex enough that it entails the potential for obstacles at nearly every step. (You better be brave. LOL !) Nevertheless, it was completely possible to always push forward because of the Metis setup: an incredibly active Slack channel (for posing and answering questions and sharing code and hyperlinks), the availability of TAs (for help via screen sharing), and an increasing sense, as the bootcamp progressed, of where and how one could seek and find (or build) solutions on one’s own.
One of the highlights of bootcamp was when the architect of Metis’s curriculum (the Metis mastermind herself) Sophie Searcy came in to guest-teach us on the topic of Singular Value Decomposition (a subtle topic of Linear Algebra that happens to be important for certain aspects of Data Science/Machine Learning). Another highlight was when Kimberly Fessel (who was _not_ one of my cohort’s instructors) came in at the end to give us (four at a time) a mock-technical interview, gently quizzing us on how to create from scratch a classification algorithm we had learned two months prior.
The variety of instructors had different emphases, different strengths, and sometimes didn’t completely agree, which is realistic, and gives a better sense of the actual kinds of divergence you’ll encounter out there in the DS world.
The instructors were great:: intense and serious about the subject matter, but also capable of being light-hearted at the right times. My main instructors were Vinny Sanguttuvan and Leon Johnson. Leon’s lectures were clear, with a definite sense of priorities distinguishing the overarching point from the supporting detail. Leon also provided excellent practical critique on bootcampers’ presentations, things that I still contemplate and consider vividly five months after bootcamp ended. Based on the particulars I had achieved with/through a given project, Leon was unique in suggesting next steps built on these and directions I could take that would further burnish my portfolio. Leon obviously has a firm overview of the DS landscape, is clear communicator, and encouraging mentor. Vinnie Sanguttuvan is one of the most dynamic and insightful teachers I’ve ever had (comparing to dozens of other great teachers I’ve had in college, grad school, as well as tech training courses). His teaching kept me and other bootcampers on our toes from start to finish. Nothing was ever routine or rote, but rather presented us challenges we had to carefully consider. Sometimes Vinny did this through Socratic-method, where he would pose a series of questions (or mini-problems) that would interactively lead to an insight, which therefore would stick. Other times he would relate mathematical/algorithmic/statistical principles to observations he had personally made about well-known tech industry products and services. When Vinny was teaching, often you really had to think through mathematical or algorithmic concepts, and problem-solve on the fly. He is also very open minded about what types of problems might be tackled with algorithmic methods. Vinny’s teaching presents a distinctive synthesis of creativity and precision that epitomizes what is most exciting about the Data Science field.
I want to emphasize that while Metis’s excellent curriculum is one of the reason’s I’m glad I chose it, the most important learning events for me were the moments when the instructor cooked up his/her own problems or reflections for us to consider; these complemented the pre-cooked curricular material. That is to say, there was a mixture of very carefully planned out lessons and more spontaneous ones originating from the instructor’s creativity. It made for an exciting and engaging bootcamp, when you could sense that what and how you were learning was unique to that moment and context. (This was especially true with approaches to the pair-programming problems). I hope that Metis continues to allow seasoned instructors latitude in how they teach, including creating their own material from time to time.
Another instructor I want to mention is Dimitri Theoharatos, who taught python-and-math course I took a year before the bootcamp. Dimitri moved quickly and efficiently, while also overflowing with a wealth of practical tips and laser-like precision. He peppered his teaching with astonishingly apropos industry insights of his own, giving me perspectives that sill condition my thinking a year later. I learned as much from carefully observing how Dimitri navigated the nuts-and-bolts of the python programming interface (jupyter notebook) as from what he verbally articulated.
All the Metis instructors listened carefully to students’ questions. They seemed to appreciate they were dealing with many intelligent and motivated bootcampers, so they calibrated their responses accordingly, never resorting to generalized rote answers (as sensitive teachers might be prone to do).
The camaraderie among bootcampers was inspiring and encouraging. One thing that was incredibly clear after bootcamp ended: Metis alumni help other Metis alumni, especially Metis alumni from prior cohorts. In the brutally frigid world out there of cold applications to jobs, the Metis alumni network is indispensable; it’s how I got most of my interviews, including the ones that led to me landing two simultaneous job offers.
Especially if you’re coming from an academic career, or this is your first job search, the Metis careers team (I worked mostly with Ashley) can be helpful in steering you what might be unfamiliar terrain of the corporate career landscape.
Before bootcamp: To make the most out of this experience, I recommend you prepare extensively. Try to use Codecademy and Dataquest to learn Python and SQL before the bootcamp. Use Brilliant.org to learn basics of linear algebra, probability, stats, and calculus. (If you’re going to splurge on the bootcamp you might as well purchase memberships to these so that you begin the bootcamp with your best self. I used these further _after_ bootcamp as well.) As many others mention in their reviews (of Metis and other bootcamps) the amount of material coming at you in the bootcamp is like a firehose to your mind, which you can absorb better to the extent you prepare in advance. If you get your feet wet with Brilliang.org, Codecademy, DataQuest, and the Metis python-and-math, you will be ready even if you don’t come from a thoroughly STEM background. During bootcamp, expect to spend all evening every weekday as well as a large chunk of weekends working on your bootcamp projects. Metis DS bootcamp is not a side activity; it’s a life-changing process.