How to Transition from Engineering to Data Science
AI for engineers: experience of an engineering graduate
Hi there,
I often get asked about my transition from engineering to data science. Having made the leap, I’ve learned what works (and what doesn’t) and I thought it would be of value to share my insight to hopefully save you a lot of time.
I have a Master’s degree in Engineering (Material Science) and an Engineering Doctorate. I am self-taught in data science and have been practicing it for 5 years. Whether you’re looking to:
- Transition into a data role
- Leverage data science knowledge for your engineering work
- Or learn these skills out of curiosity
...this guide is for you. It’s based on my journey and will help you navigate your transition into the world of AI and data science with confidence.
Why engineers excel in data roles
Engineers have all the tools to be great machine learning practitioners. They are problem solvers, and they tend to be practical. They can cut through the noise and leverage what already works to build a solution to a problem. Without getting bogged down trying to code a new neural network from scratch.
In addition, engineers have a lot of “domain” (subject-specific) knowledge in engineering. Data scientists from other backgrounds may lack this knowledge, which gives engineers an advantage in all engineering-related data and machine learning tasks.
But for many engineers, AI can feel daunting and out of reach.
So — how to get started?
Hands-on first, theory later.
Forget theory-heavy textbooks — you must start hands-on. You’ll learn better and faster by doing. I tried to learn machine learning 3 times. The first two times, I started with online courses packed with math-heavy theory first- how neural networks work, gradient descent, loss functions, etc. Both times, I quit within a month. Why? I wasn’t applying what I was learning.
On the third try, I took a practical course that had me coding from Lesson 1. Problem-solving right away. I was hooked. My brain loved the instant feedback and application of my new skills. And here I am 5 years later still coding and solving data problems.
The theory behind the algorithms is important, but you can always learn it later when you need to. The top-down approach is hard for some to take. But for most, learning the theory first just kills the initial joy and motivation.
I find that the hands-on skills are going to serve me much better in the long run than getting a certificate for watching a few lectures.
Practical roadmap: all roads lead to Kaggle
This brings us to Kaggle. A practical, fun, free website known for hosting machine learning competitions with big prize pots. Kaggle also provides free datasets, a coding environment, notebooks (code solutions) shared by others, and most importantly for us — free hands-on courses on coding and machine learning.
Kaggle got me hooked on my third attempt at learning machine learning. And it’s still the platform I recommend to engineers looking to get started in AI. And every engineer who took up my advice has sworn by Kaggle since. It works because it’s all about solving problems right away. You can find the free Kaggle courses here.
This is the order in which I would take these mini-courses if I were to start over:
· (If you haven’t programmed before) Intro to Programming
· Python
· Intro to Machine Learning
· Pandas
· Intermediate Machine Learning
· Data Cleaning
· Feature Engineering
· Machine Learning Explainability
It starts with Python. Forget MATLAB. Python. Python allows you to use the latest algorithms out of the box, is free, well documented and offers an accessible way of applying machine learning in your daily work.
With the basics of Python learned, it’s time to build your first models and really get hooked on machine learning before you lose the motivation! “Intro to Machine Learning” will let you build your first models and join your first machine learning competition to help gamify your learning process and fuel your motivation further.
Going forward all the other recommended mini-courses will help build your foundation. Kaggle does offer other courses, and I do think they are all valuable and worth reviewing once you get the basics down.
Now, if you make it this far and are still interested, this would be the point I would invest in my first book. Only 1 recommendation is needed here and that’s Hands-On Machine Learning by Aurélien Geron. It’s practical, goes further than Kaggle on each topic, and can serve as a great reference point for your future projects. (I earn a commission if you buy through this link, at no extra cost to you.)
For most engineers transitioning to data science, foundational knowledge of machine learning is sufficient to get started. Advanced AI concepts like deep learning or transformers do become important when tackling problems such as image recognition or natural language processing. However, I would only delve into these areas after mastering the basics and facing projects that demand this knowledge. Chances are, most of the problems you’ll work on will be tabular or time series-based, as much of the data you’ll encounter will come from sensors and logs.
Focussing on what will get you the most results can make the process less daunting and more goal-oriented.
Applying learned skills
Once you’ve completed foundational Kaggle courses, it’s time to shift from solving generic problems to addressing domain-specific challenges. The majority of the problems you would have worked on so far would likely not be super relevant to the problems you will be working on going forward.
At this stage, it’s important to start showcasing your new skills by tackling problems relevant to the domain you plan to work in. Start by defining that domain, whether that's engineering or something else entirely. With engineering, you can narrow down further to your specific niche — like materials, machining, etc. And then you can find some problems to solve within that niche.
A good place to start is to also look for projects in your workplace where inefficiencies could be reduced or outcomes improved with a machine learning solution. You can then present a small proof-of-concept model to your manager, which might spark interest and open up more opportunities to integrate data science into your role. A lot of companies (especially engineering ones) prefer upskilling internally to hiring a dedicated data scientist, as they may not have enough workload to justify a full-time hire. This can serve you perfectly during your transition.
If you are stuck for ideas, I’m sure ChatGPT can help you brainstorm. Predictive maintenance, quality control and logistics optimisation are all examples of engineering use cases of AI.
In my case, I began by identifying use cases and inefficiencies in my workplace. The very first project I worked on already had data collected and had a solution built which relied on using linear regression to predict part condition after machining. I took the dataset and passed it through a classic data science workflow. I cleaned the data, visualised it, did some feature engineering and feature selection, built a baseline, tried multiple models, evaluated each and optimised the best-performing one.
These projects help you build some confidence while also growing your portfolio, which will eventually open doors to data science roles. A useful tip here is that presentation is everything. Make sure the problem is clearly defined and the results are communicated well in a business context. If the model can reduce costs or improve efficiency — emphasise it.
Hands-on machine learning for engineering
If you’re interested, I’m also developing a hands-on course designed for engineers. The course focuses on one of the most impactful areas I’ve worked in for the past 5 years: predictive maintenance—using AI to prevent equipment failure. Designed with real-world datasets and case studies, it requires no prior Python knowledge and teaches everything I wish I had when starting out.
For a limited time, the course is available at an exclusive pre-sale discount:
Final takeaway
Just get started and have fun! Stay curious, keep solving problems, and you’ll carve out your place in the exciting world of AI.
And if you find that machine learning isn’t your thing after trying it out, you’ll still gain valuable knowledge that’s increasingly relevant in today’s world.
I’ll be sharing more data science projects and insights in future posts-subscribe to stay updated!
ML in Motion | Dan Pietrow | Substack
Originally published at https://danpietrow.substack.com.
How to Transition from Engineering to Data Science was originally published in Towards Data Science on Medium, where people are continuing the conversation by highlighting and responding to this story.
from Datascience in Towards Data Science on Medium https://ift.tt/U2ADK38
via IFTTT
