Two lessons from hundreds of Machine Learning interviews
I’m a Machine Learning Engineer (MLE) with 5 years of experience. I’ve recently interviewed for a dozen big companies in BayArea. I also interviewed 100 other people over the years for Machine Learning engineer positions. There are two particular challenges for Machine Learning Engineer interviews.
Know your priority
After I went through the interview process, I designed a few Machine Learning quizzes to help people test their skills. The tests are focused on many areas: coding, data science, and SQL. Surprisingly, 40% of people couldn’t answer DecisionTree or RandomForests basic questions (see chart). I have discussed it with some of them and noticed a pattern. Some people focus on the latest and greatest Deep Learning advancements, others focus on Leetcode programming style questions. It was quite clear that deciding which area to focus on is guesswork for many people.
There are plenty of books for Software Engineering Interview: Cracking the Coding Interview, The Element of Programming Interview, and Leetcode. They provide a guideline with a clear focus area for planning their study. For machine learning engineer interviews, there are no popular books or guidelines. Without a clear structural guideline, it’s very easy to chase the wrong paths.
Machine Learning system design is hard
This is the famous diagram in “Hidden Technical Debt in Machine Learning Systems” paper by D. Sculley et al.Google. In a production system, the Machine Learning model is one small part among many other components. It’s hard to design ML systems without knowing how it fits in the whole system. Another example is the inference latency requirement in a Machine Learning system. Machine Learning engineers have to balance between model performance and model inference latency.
In Machine Learning system design, you have to pay attention to the training data generation strategy. A bad data generation strategy can lead to good offline model metrics but can lead to poor performing models with online metrics.
Pick the right formulation
One study guide for Machine Learning interviews
As I went through the whole process. I knew a better way to prepare for the MLE interview. I released my Study Guide and ML system design on my Github. It went viral on Reddit and LinkedIn. You can test your ML knowledge and compare it with others. The main purpose is to help you prepare for your Machine Learning Interview.
Additionally you can also learn Machine Learning Design course on educative.
Support me to read more content like this. If you have any questions, you can drop me an email: helppreparemle@gmail.com.
