Should I get a data science degree? (Part I)

Bro: Are you stupid? definitely go to school to study data science.

Other Bro: Yeaah…. maybe not, you should probably avoid wasting moneys’ and learn everything online. The internet ever heard of it, bro ?.

Bro: Weeeeelll, you should learn some math before calling yourself a data Scientist though, no?

And there they went for ages debating the fall of the educational system and the internet MOOC revolution versus sound, structured learning, the importance of fundamentals, and the need for evaluation to assess one’s skills…

An “unsplash study” image to make the article popup…

To cut to the chase, I did both and I’m here today to recap my own experience of learning data science using various online resources first (I’ll detail all the resources in part III of this series ), then actually going to one of the best French engineer school and getting a data science MS degree.

I’ll start with a disclaimer, I did not start from scratch i.e. no undergrad/grad school diploma and basic high school math. To briefly introduce myself, I graduated in 2017 from a French engineer school (Telecom IPP) with a master’s degree in IT (very vague for a diploma, I agree, and I’ll discuss this general diploma in a later post ). The important thing is that I had a pretty good foundation in mathematics: I had two years of highly competitive maths and physics classes before attending Telecom and had always had quite a love for mathematics. I had a good understanding of Linear algebra, calculus, and arithmetics .. due to gulag level torture from those two years.

Ok, so I had 2 years of strong maths foundation followed by 3 years of basically breezing through computer science classes and basically forgetting 90% of the courses 3 months after graduating.

Now 3 years after graduating, I’m a consultant, working 7 hours a day editing slides and spreadsheets, hating every hour of the day, and thinking: why the hell am I wasting my time doing this?

I know this is not a therapy session, but hey I’m writing.

So after trying day trading crypto, trying to pivot to the fitness industry and passing various certifications then trying to apply to a biomechanics degree. I finally stumbled upon what will come to become my holy grail: MACHINE LEARNING.

It all started talking to one of my friends, who also was as miserable as me working as a consultant. We talked about the importance of learning hard skills, why we ended up doing this kind of job, and other subjects I cannot disclose for national safety reasons … I mentioned a Coursera course on machine learning I just saw and how we could maybe start it together. Little did I know that the course was the famous Machine Learning by Andrew Ng.

I religiously worked on that MOOC, I worked every night after work, installed Matlab, read about the algorithms, and refreshed my memory to rekindle the mathematics I forgot. I studied linear algebra again, probability and statistics, calculus, and information theory. 1 month later I knew this was my journey and I was not going to stop there. I learned about regression, clustering, SVM, Mixture Gaussian Models, and recommender systems… The course had the right amount of depth and breadth I needed to not be discouraged by the field at that time. Andrew Ng, is an amazing teacher and seeing him explain at a high level the algorithms and their implementation gave me the courage to look at them more closely and refresh my mathematics when needed to understand the theory a little more closely. I’m also very curious by nature about details and diving that deep into details at this stage was not necessary to have a global vision about the landscape of ML algorithms.

Finally, on a Wednesday night, I received a certificate from Coursera. I printed it and stared at it for a moment. It represented a physical token that I do have the tools to change my life.

Ok, what’s next? I started looking at what to learn first (I knew that my why was already answered for by some deep demons). I started by creating a backlog of all the resources I gathered: MOOCs, books, articles, MIT courseware lessons, youtube lectures, and articles. I had it all mapped out on a Notion calendar. While parsing through the resources, I very carefully tried to check myself. See, I had the freedom to choose what I wanted to learn but also the responsibility to choose correctly, to start with the fundamentals and get to the sexy stuff later.

After looking on the internet and reading about data science, I knew I had to learn python to be a ‘real’ Data scientist. I loved MOOCs and knew that I am a visual learner. Coursera MOOCs and youtube videos gave me the ability to watch videos at 2x, 3x and I could advance at my own pace, pause it, forward parts I already knew, rewatch for the 10th time the same obscure proof…

The next stop was the Applied Data Science with Python taught by Michigan university on Coursera. In parallel, I read about python in separate books, did a basic project, and refreshed my memory about programming languages, programming paradigm, data structures, complexity … After this MOOC, I knew I could do some data science stuff, but I needed to apply all that freshly learned knowledge. I headed over to Kaggle to see what all that hype was about and started looking at some datasets and projects (The Titanic one of course) …. but wait what was this XGBoost magic ?? How can a model do good on a dataset and not on another? Clearly, I had perhaps overestimated my knowledge (don’t we all?) Let’s go back to learning.

Computer vision always had a special place in my heart, I always loved the challenge of achieving human-level perception using only a matrix of pixels as input. I also wanted to learn about all that Deep learning I kept hearing about and that ‘artificial intelligence’ so naturally I took the deep learning specialization taught by Andrew Ng. This course was very very good! Spoiler alert on part II: It was basically 100 times better than anything practical we did when learning deep learning in the master’s course. The course programming assignments were amazing, I wrote everything from scratch (only using Numpy as a dependency): a neural network with forward propagation and backpropagation, a convolutional network, a ResNet, LSTM, a language model… It also had some pretty good nuggets about structuring a machine learning project. I catch myself going to that section of the course now and then. I always felt like I had to keep learning until I could deeply understand a research paper (mind you any paper…). It always felt like something important to achieve. So after finishing this course I had read the majority of seminal papers in computer vision and got the majority of it! Huge win!

I felt that I learned enough tools to be a real ‘data scientist’. I quit my consulting job right at the same time as the first lockdown in France in March. I wrote a medium article at the time using that fresh LSTM knowledge to get a pessimistic upper bound on the Covid numbers in France. In this article, I made it out pretty clear that this period was a pretty unique opportunity, even citing Newton and what not. Putting my money where my mouth is, I started a startup with my friend. The first fitness app developed to share a video call workout session without ever leaving the comfort of your home: Fitroulette. I dove into web and mobile development, networking, WebRTC, React native and you guessed it: Deep learning. I worked on a pose estimation module and a sports language model. It was a good learning experience. We had a beta version working in August of 2020.

It was time to make a choice and the Covid situation was a little bit tricky. Right after quitting my job, I tried finding a data science position by only presenting my portfolio of projects and the Coursera certificates (and also a Telecom diploma to be honest), most rejected me because I didn’t have enough experience in Data Science. I applied to some Data science masters and my thinking at the time was: “I pretty much know this stuff, I’ll get a diploma while working on my startup and finally get a piece of paper for employers as a safety plan”. Little did I know what this journey would look like…