🤡 How to Fail a Technical ML Interview in 3 Minutes

A candidate comes in for a junior or even mid-level position. They ask about their tech stack — it's a full spread, the person has practically trained their own Claude on home GPUs.

They give them a simple linear regression problem. The person happily imports scikit-learn, writes .fit(), then .predict(). Takes exactly 20 seconds. They're pleased.

Then I start asking questions:
— Why did you choose MSE for quality evaluation instead of R2?
— We have a large outlier in the dataset. How will your model react to it and why?
— What does that coefficient for feature X mean if we haven't normalized it?

And silence... The person knows the library API but has no clue how everything works under the hood.

Tools like sklearn were created by smart people for smart people to save time on routine tasks. Not so that someone can hide a lack of fundamental knowledge behind two lines of code. The algorithm doesn't care about your data; it will devour any garbage you feed it and give you a prediction. Garbage in — garbage out.

These are the basics we will start studying in the next season of the Master Group. The program includes:
🟢 5 live sessions: from the basics of machine learning to building and validating basic linear models.
🟢 1 session — team practice. Enough just watching lectures. You'll work in teams on a practical task, and then we'll all review the results together (and see who messed up where).
🟢 Recordings and my code files, of course, stay with you. Take them and use them as a cheat sheet at work.

🗓 Reminder that we start on April 25 at 11:00 Moscow time. You should already know Python and pandas.

👉 If you want to stop being a fit-predict monkey — write to me directly at @obulygin91, I'll tell you the details.