It was initially painful but ultimately rewarding. I was lucky to have worked with an amazing team and learned a ton. Du

Think about it this way: each person has a relatively constant skill capacity that can be spread out to different topics. Traditionally, people’s skillsets form a software engineering cluster and a modeling cluster. As the space gets more complex, the boundary between these two clusters doesn’t make sense anymore. A person can be deeply specialized in a collection of topics in both clusters. The total volume they cover in this space is still roughly the same as traditional university-educated people, but the possible combinations of topics and depths increase significantly.,Having worked as a software and machine learning engineer for several years, I have developed my philosophy of learning in this field. I decided to write about my experience and share some advice with newcomers on what not to do, and what to do instead.,The wrong way to start is to try to cover as many points as possible in that subway map. In addition, if you want to become end-to-end, you need a similar map but for software engineering as well! A sane person would immediately see through the complexity on the surface and identify the real question:,I’ve chosen four experiences that I am aware of (from both practice and psychiatric studies) which reliably carry a host of negative emotional consequences. Obviously, life presents all manner of heartbreaking episodes. But there are some events that bring with them more than just challenge and can overpower most people’s coping abilities. The end result is emotional turmoil. For those with even the slightest vulnerability to mental disorders, this descent may tip the balance toward becoming symptomatic.,Nowadays, it is not unreasonable to ask for both software engineering and machine learning skills from one person. Since both are very big fields, it is unreasonable to expect deep knowledge in every domain, but the person should have a functional set of skills where they can handle end-to-end workflows in their strongest area reasonably well.,To be more in-demand in the current job market by being end-to-end, and not subject yourself to the limits imposed by traditional educational systems, you need to be result-driven. Recognize your personal interests and the trend in technology, create our own path that leads to value creation.,The field of data science and machine learning engineering is enormous and fast-evolving, nobody can learn everything. Traditional universities are too slow to design new curriculums. At the same time, there are tons of world-class educational resources online. The advantage of designing such a customizable online degree is easy to see. Even so, it is very hard for beginners to design such a degree for themselves because what they often see is stuff like this:,With a short period of intense preparation and some luck, I started my career as a software engineer at a Silicon Valley company, working on full-stack web development with some of the top engineers in the field.,I went into the office every day feeling extremely insecure, grinding my way through each line of code, and somehow made it work without understanding the big picture how each component talked to each other.,This is not trying to be a one-size-fits-all machine learning degree. There are different career paths in machine learning. Everyone has their own background, career goals, and interests. The diversity in the field is healthy and should be encouraged. What shouldn’t be encouraged is the wrong organizational separation of responsibility. But more on that later.,Earlier this year I came across a viral Twitter thread by Randall Kanna about how to create one’s own computer science degree with free online content. It was not only excellent for people who don’t have prior knowledge in computer science, it’s also valuable for new software engineers who didn’t major in CS in college. That’s when I thought about creating my own full-stack machine learning engineering degree. After being in the tech industry and having learned it the hard way, I believe a custom-designed curriculum is going to be valuable to myself and others who have similar goals like mine. I curated a collection of the best resources I found, wrote down a curriculum, shared it on Github and Twitter where I called it “My CS Degree” with a focus on Full-Stack ML Engineering. Since then, a lot of people have starred and forked it. I’m glad that they found it helpful.,Why the emphasis on “end-to-end” and “production”? Some data scientists build models and hand them off to engineers to productionize later. This is how data science organizations operate in many companies now and it is gradually recognized as an anti-pattern for those who need models in production. In Eugene Yan’s excellent article Unpopular Opinion — Data Scientists Should Be More End-to-End, he explained why it is the case eloquently. I completely agree and won’t repeat it here.,One major reason that this separation exists is the distribution of expertise in the talent market. Internally, it becomes perceived expertise based on job titles and descriptions. The average data scientist has a degree in statistics or other math-heavy fields with less knowledge in computer science. An engineer usually has a CS degree but lacks modeling expertise. Do you see the difference? It is a legacy issue caused by traditional universities. For most companies, there aren’t enough talents who learned both well enough in school or from previous experience. Even if the talents exist, a wrongly structured organization tends to underutilize them, or even pit data scientists and engineers against each other. It is time to rectify it with a set of more up-to-date expectations.,In this article, I will explain the philosophy behind this curriculum, who can benefit from it, and how to reason about what and how to learn given your personal goal and the exploration-exploitation dilemma in the vast world of knowledge. If you are only interested in the actual curriculum, check it out on Github.,A little digression before I dive into the structure and philosophy of this degree. I have an undergraduate degree in physics and used to study machine learning in grad school before deep learning became the hottest thing. I didn’t study computer science systematically but did take courses in algorithms and data structures. So you may see me fall into the “data science” camp mentioned above in terms of education. As for real-world software engineering and web development, I was clueless when I graduated.