Machine Learning · Module 2Module material
Exploring a Dataset for Modelling interview questions
12 questions interviewers ask on this module’s topics, from first principles to the follow-ups. Each comes with a model answer.
Signal and redundancy
- How do you tell whether a feature carries information about a binary label, before training a model?●
- Why not rank features by the difference in class means?●●
- What is the relation between the CDF gap and a threshold's accuracy?●●
- What is a redundant feature, and how do you find one?●
- Two features correlate at 0.787. Can you drop one?●●
- Why look at a scatter when you already have the correlation?●
Leakage and codes
- What is target leakage? Give an example.●
- How do you find leakage in a table you have not seen?●●
- What is a sentinel value, and why is it dangerous?●●
- A categorical column is
unknownfor a fifth of the rows. What do you do?● - A feature is known before the prediction and predicts very well. Can it still cause a problem?●●
- Why does accuracy barely move for a feature with a strong signal?●