Learning was difficult.
There were many courses, books and resources I used along the way that helped me, but honestly, I wouldn't have taken many of them in hindsight.
So I would like to review everything I've learned to do my job to land work with machine learning. And I would like to tell you which areas were actually valuable and which areas were not.
Let's get into it!
University degree/mathematics
I am extremely fortunate that I decided to study for my master's degree in physics as a teenager.
Yes, you're probably rolling your eyes right now.
“This guy said he doesn't have a background in CS, but he said, Master's degree in physics, what a goddamn.”
There is no denying that this definitely gave me an advantage. However, many STEM alumni still struggle to find a machine learning job. I worked with them.
Simply putting a master on a STEM subject is far from guaranteeing you can easily get into the job.
There is much more to learn, but this is not usually taught in most programs.
That said, the main thing I learned in my current degree related to my work as a machine learning engineer was mathematics skills.
I've learned calculus and linear algebra to a fierce level, rather than you need to be honest, and need statistics for a decent standard. Still, I had to refine my knowledge of statistics later.
It was my first time writing code for my degree.
Literally, my first day, at 9am, I had a computer lab tutorial Fortran.
For those unfamiliar, Fortran is the oldest “high-level” programming language invented in the 1950s. But here we were taught that in 2017.
Fortran is hardly for beginners and I didn't like programming right away. As I got older I knew what I was doing today!
I didn't enjoy Fortran, but in the long run it taught me how to think about and solve problems using a code that paid dividends.
If you want to know all the math skills you need to work with machine learning, check out our previous post.
How to learn the mathematics needed for machine learning
Breakdown of three basic mathematical fields required for machine learning: statistics, linear algebra, and…medium.com
Python
I hated Fortran so much that I actively avoided modules with programming aspects.
However, in 2020, the video was recommended on the YouTube homepage in its third year.
Alphago – Movie
For those who are not aware, this was a documentary about the AI Alphago of Deep Mind, who defeated the best go-player in the world. Most people thought GO wasn't good at, let alone AI defeating the world champions.
After watching the video, I began to read about how AI works, including neural networks, reinforcement learning, and deep learning.
Since then I've been obsessed with being a data scientist and I knew I had to learn Python to become Python.
In the evenings and weekends, I go through several Python courses and projects. Here's what I used:
Not to mention the endless Google search and Stackoverflow threads I visited. After all, this was before Preshatogput.
I also practiced Python skills. Hakke rank Building basic projects for problems and fun and my university coursework.
SQL
After studying Python, I devoted about a month to studying SQL, applying for entry-level and graduate data science jobs.
SQL is easier to learn than many other languages, and the basics are that it covers almost anything you want to do.
The courses and resources used for SQL are as follows:
And again, I used it Hakke rank Practice SQL problems for interviews.
This is just a part of my learning journey and I have mastered most of my advanced SQL skills at work.
Machine Learning
In my last year in college, I took it Andrew NG's specialization in machine learning. I took it when it was still the 2012 version and the coding exercise was octave/MATLAB.
This course taught us the theoretical foundations of all machine learning algorithms, including:
This was everything before I started implementing it in my code. Building that intuition behind an algorithm is invaluable.
We also supplemented our learning with various textbooks.
All of these are still used today. Because you will forever study and update your knowledge about machine learning.
Deep learning
After studying all the basic machine learning knowledge, I took subsequent courses by Andrew Ng. Deep learning specialization At Coursera.
I once again complemented my learning with the same textbook as the machine learning section to cover many advanced concepts.
Here are more videos and courses I used:
statistics
At this point in my journey, I got my first job as a data scientist at an insurance company. There, I worked closely with the actuaries.
For those who don't know what an actuary is, Wikipedia I'll explain them:
Actuaries are experts with advanced mathematical skills in dealing with the measurement and management of risk and uncertainty.
I have studied statistics before, but the level required for insurance companies is relatively high, especially when working with actuaries, especially as they are on-site experts.
I did some research to upgrade the statistics CS1 (Statistics) Actuarial exam. I hadn't actually had an exam, but I did review and studied everything.
The syllabus covers most of all statistics that you may use as a data scientist or machine learning engineer throughout your career.
Books Practical statistics for data scientists (Affiliate link) Provided as reference text to update my knowledge, I studied Think Bayes (Affiliate link) A textbook for learning Bayesian statistics.
It is important to note that I simply did not take the course and read the book. I have essentially documented everything I learned in the medium.
General statistics
Probability distribution
Bayesian statistics
As I have said many times, this was the biggest ROI of my career.
Time Series Prediction
After spending a year on insurance, I switched companies and worked for a team dedicated to time series forecasting and optimization issues.
The only book I used to learn predictions was Prediction: Principles and Practice (Affiliate link) Rob Hindman and George Atanasopoulos.
This is known as the Bible of Prediction and is the only book that people recommend getting when they start studying the field.
The remaining knowledge obtained online from Google searches and random videos. This was usually a way of complementing my knowledge in most fields.
And of course, I documented everything in the media.
Time series
Optimization/Operational Research
My knowledge of optimization was a bit more mixed because it's a vast field. To give you a sense of size, it definitely covers all machine learning, and Discrete Optimization Algorithm.
The main reference text I used was Algorithms for Optimization (Affiliate links), And I've complemented it with a variety of other online resources, such as:
But in general, I study areas where I need to learn for my work and write blog posts about them. That's what I learnt most of and honest, and still.
Software Engineering
When I was considering moving from a data scientist to a machine learning engineer, a key area that I needed to improve was software engineering skills.
It's a big area and in fact it's the whole job, but I focused on the basics.
The course I took is as follows:
One area that is difficult to investigate is creating the right production code. This is the only thing I've learned from my job, but creating my own software projects allows me to gain experience outside.
If that seems like there are a lot, don't worry.
Also, like I said at the beginning, not all of it was necessary in hindsight. The next area is something I definitely won't do again.
- Actuarial CS1 – In reality, you don't need many concepts. The mathematical details are overkill. I recommend sticking to Practical statistics for data scientists (Affiliate link)textbook.
- CS107 Computer Organizations and Systems-I haven't used much ideas here.
- Elements of Statistical Learning –Excessive textbooks for most people.
The rest was definitely worth it, but I definitely didn't need all of these resources. One good thing in each section is enough.
If you are after a proper and detailed roadmap for breaking into machine learning, we recommend checking out the previous post below.
The ultimate AI/ML roadmap for beginners
How to learn AI/ML from scratch
One more!
We offer 1:1 coaching calls that let you chat about what you need, whether you're just figuring out what you need, whether you're grasping the next step or not. I'm here to help you move forward!
1:1 Mentoring call with Egor Howell
Career guidance, job advice, project help, resume review
